diff --git a/.obsidian/graph.json b/.obsidian/graph.json
index 1bd25ab..1246aa7 100644
--- a/.obsidian/graph.json
+++ b/.obsidian/graph.json
@@ -17,6 +17,6 @@
   "repelStrength": 10,
   "linkStrength": 1,
   "linkDistance": 250,
-  "scale": 0.45971825916974024,
+  "scale": 0.7108694604315626,
   "close": true
 }
\ No newline at end of file
diff --git a/99 Daily/2026-04-15.md b/99 Daily/2026-04-15.md
index 022674a..c70faef 100644
--- a/99 Daily/2026-04-15.md	
+++ b/99 Daily/2026-04-15.md	
@@ -14,3 +14,15 @@ tags: [daily]
 - 10:27 — Claude Code session ended | `memory-compiler`
 - 10:32 — Claude Code session ended | `memory-compiler`
 - 10:33 — Claude Code session ended | `memory-compiler`
+- 10:40 (1min) — Claude Code session ended | `aimpress`
+- 10:41 — Claude Code session ended | `aimpress`
+- 10:42 — Claude Code session ended | `aimpress`
+- 10:46 | `aimpress`
+  - **Asked:** Remove Ollama and all associated files and data.
+  - **Done:** Uninstalled Ollama, dependencies, models (6.1 GB), configs, logs, and background services.
+- 10:46 | `aimpress`
+  - **Asked:** Set up an Obsidian logging system that automatically captures all Claude Code interactions across projects with proper session documentation.
+  - **Done:** Identified that the Stop hook needs to be rewritten to parse transcripts and call Haiku for structured entries, rather than just logging timestamps.
+- 10:47 | `aimpress`
+  - **Asked:** Set up a system to automatically log all Claude Code interactions across projects into Obsidian with proper structure and formatting.
+  - **Done:** Configured structured logging format that captures "Asked" and "Done" for each session, tested it on current transcript, and verified it works with Haiku-powered processing (~$0.001 per session).
diff --git a/wiki/_master-index.md b/wiki/_master-index.md
index fa076f9..91e667c 100644
--- a/wiki/_master-index.md
+++ b/wiki/_master-index.md
@@ -19,6 +19,10 @@ This 3-hop pattern works for hundreds of articles without vector search.
 | Topic | Description | Articles |
 |-------|-------------|----------|
 | [[wiki/obsidian-rag/_index\|obsidian-rag/]] | Karpathy's LLM wiki method — Obsidian RAG, setup, vs true RAG | 3 |
+| [[wiki/projects-overview/_index\|projects-overview/]] | All 36 Oliver Agency projects — navigable table with stack + server | 1 |
+| [[wiki/tech-patterns/_index\|tech-patterns/]] | Recurring tech stacks: FastAPI, React/Vite, Next.js, Azure AD, AI, Box, One2Edit | 9 |
+| [[wiki/architecture/_index\|architecture/]] | Cross-cutting architectural patterns: Docker Compose, multi-agent AI, GCP timeout, RAG, hotfolder | 5 |
+| [[wiki/client-knowledge/_index\|client-knowledge/]] | Per-client notes for Ford, H&M, L'Oréal (2+ projects each) | 3 |
 | [[wiki/concepts/_index\|concepts/]] | Atomic knowledge extracted from Claude Code sessions | 0 |
 | [[wiki/connections/_index\|connections/]] | Cross-cutting insights linking 2+ concepts | 0 |
 | [[wiki/qa/_index\|qa/]] | Filed answers to queries (saved with `--file-back`) | 0 |
diff --git a/wiki/architecture/_index.md b/wiki/architecture/_index.md
new file mode 100644
index 0000000..f3067f4
--- /dev/null
+++ b/wiki/architecture/_index.md
@@ -0,0 +1,29 @@
+---
+title: "Architecture Patterns Index"
+description: "Cross-cutting architectural decisions across Oliver Agency projects"
+tags: [index, architecture]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# Architecture Patterns
+
+Cross-cutting architectural decisions that appear in multiple Oliver projects.
+
+## Patterns
+
+| Pattern | Description | Projects |
+|---------|-------------|---------|
+| [[wiki/architecture/docker-compose-fullstack\|docker-compose-fullstack]] | Standard multi-service Docker Compose deployment | ~20 projects |
+| [[wiki/architecture/multi-agent-ai-systems\|multi-agent-ai-systems]] | Parallel specialist agents + lead synthesizer | Mod Comms, Semblance, Enterprise Nexus |
+| [[wiki/architecture/gcp-deployment-lb-timeout\|gcp-deployment-lb-timeout]] | GCP 30s LB timeout — WebSocket → HTTP polling fix | Mod Comms, Semblance |
+| [[wiki/architecture/rag-architecture\|rag-architecture]] | RAG: Firecrawl → AI structuring → Qdrant → LLM synthesis | Enterprise Nexus, Sandbox NotebookLM |
+| [[wiki/architecture/hotfolder-daemon\|hotfolder-daemon]] | Box folder monitoring daemon with systemd | Ford QC, Ford SFTP |
+
+## Key Architectural Decisions
+
+1. **Docker Compose** — default deployment for all multi-service projects
+2. **HTTP polling over WebSocket** — mandatory on GCP (30s LB timeout)
+3. **AI pre-structuring before RAG indexing** — improves retrieval quality
+4. **Hotfolder + archive pattern** — prevents reprocessing in Box automations
+5. **DEV_AUTH_BYPASS** — skip Azure AD in local dev, always use real auth in production
diff --git a/wiki/architecture/docker-compose-fullstack.md b/wiki/architecture/docker-compose-fullstack.md
new file mode 100644
index 0000000..289f65d
--- /dev/null
+++ b/wiki/architecture/docker-compose-fullstack.md
@@ -0,0 +1,128 @@
+---
+title: "Docker Compose Fullstack Deployment"
+aliases: [docker-compose, docker, deployment, devops]
+tags: [docker, docker-compose, deployment, fullstack]
+sources: [01 Projects/gmal-scope-builder, 01 Projects/modcomms, 01 Projects/enterprise-ai-hub-nexus, 01 Projects/video-accessibility, 01 Projects/sandbox-notebookllamalm-nextjs]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# Docker Compose Fullstack Deployment
+
+The standard Oliver deployment model. Every service (frontend, backend, DB, Redis, workers) runs as a Docker Compose service.
+
+## Key Takeaways
+- `docker compose up --build` is the standard single-command deploy for ~20 projects
+- Pin port to non-default for PostgreSQL (5433, not 5432) to avoid local conflicts
+- Backend and frontend build separately — rebuild only what changed to save time
+- Always copy `.env.example` → `.env` before first run
+- `deploy.sh` script in some projects (GMAL) automates env setup + build
+
+## When to Use
+Any project with 2+ services (e.g., API + DB, or frontend + backend + DB).
+
+## Standard Compose Patterns
+
+### 3-Service Stack (Most Common)
+```yaml
+# docker-compose.yml
+services:
+  frontend:
+    build: ./frontend
+    ports: ["3000:3000"]
+    depends_on: [backend]
+
+  backend:
+    build: ./backend
+    ports: ["8000:8000"]
+    env_file: .env
+    depends_on: [db]
+
+  db:
+    image: postgres:16
+    ports: ["5433:5432"]   # non-default external port
+    volumes: ["pgdata:/var/lib/postgresql/data"]
+    environment:
+      POSTGRES_DB: mydb
+      POSTGRES_USER: myuser
+      POSTGRES_PASSWORD: mypass
+
+volumes:
+  pgdata:
+```
+
+### Full AI Stack (Enterprise Nexus)
+```yaml
+services:
+  frontend:   # Next.js 14
+  backend:    # FastAPI + Celery
+  postgres:   # PostgreSQL 16 (nexus-postgres)
+  redis:      # Redis 7 (nexus-redis)
+  qdrant:     # Qdrant vector DB (nexus-qdrant)
+  worker:     # Celery worker
+  beat:       # Celery beat scheduler
+```
+
+### With nginx Proxy (DeckForge)
+```yaml
+services:
+  nginx:      # :80 → routes to frontend or backend
+  frontend:   # Next.js :3000 (internal)
+  backend:    # FastAPI :8000 (internal)
+```
+
+## Dev Commands
+```bash
+# Standard workflow
+cp .env.example .env         # First time only
+docker compose up --build    # Build + start all
+
+# Faster iteration
+docker compose build backend && docker compose up -d backend
+docker compose logs backend --tail=50
+
+# DB operations
+docker compose exec db psql -U myuser -d mydb
+docker compose exec backend alembic upgrade head
+
+# Backup DB
+docker compose exec db pg_dump -U myuser -d mydb > backups/dump.sql
+```
+
+## Production Deploy (Sandbox NotebookLM Pattern)
+```bash
+# On server (optical-web-1)
+git pull origin main
+docker compose build backend   # or frontend or both
+docker compose up -d
+docker compose logs backend --tail=50
+```
+
+## Port Conventions
+| Service | Internal Port | External Port |
+|---------|--------------|---------------|
+| FastAPI backend | 8000 | 8000 or 8001 |
+| React/Vite frontend | 3000 | 3000 or 3010 |
+| Next.js | 3000 | 3000 |
+| PostgreSQL | 5432 | **5433** (avoid local conflict) |
+| Redis | 6379 | 6379 |
+| Qdrant | 6333 | 6333 |
+
+## Projects Using This Pattern
+- [[01 Projects/enterprise-ai-hub-nexus/Enterprise AI Hub Nexus|Enterprise Nexus]] — 7-service compose (frontend, backend, postgres, redis, qdrant, worker, beat)
+- [[01 Projects/gmal-scope-builder/GMAL Scope Builder|GMAL Scope Builder]] — 3-service (frontend :3010, backend :8001, db :5433)
+- [[01 Projects/modcomms/Mod Comms|Mod Comms]] — 3-service, deployed to GCP
+- [[01 Projects/video-accessibility/Video Accessibility Platform|Video Accessibility]] — multi-worker compose with Celery + MongoDB
+- [[01 Projects/sandbox-notebookllamalm-nextjs/Sandbox NotebookLM|Sandbox NotebookLM]] — Next.js 15 + FastAPI + PostgreSQL 18 + Redis 7
+
+## Gotchas & Lessons
+- Don't use port 5432 externally for PostgreSQL — conflicts with local Postgres installs
+- `.env.example` → `.env` step is almost always required; document it in README
+- `docker compose up --build` every time is safe but slow — use targeted `build backend` for iteration
+- Celery workers need their own container with separate `command:` override
+- Volume names should be prefixed with project name to avoid sharing data between projects
+
+## Related
+- [[wiki/tech-patterns/fastapi-python-docker|fastapi-python-docker]] — the backend being deployed
+- [[wiki/tech-patterns/redis-celery-worker-queue|redis-celery-worker-queue]] — worker services
+- [[wiki/architecture/gcp-deployment-lb-timeout|gcp-deployment-lb-timeout]] — GCP-specific constraints
diff --git a/wiki/architecture/gcp-deployment-lb-timeout.md b/wiki/architecture/gcp-deployment-lb-timeout.md
new file mode 100644
index 0000000..4a38f42
--- /dev/null
+++ b/wiki/architecture/gcp-deployment-lb-timeout.md
@@ -0,0 +1,83 @@
+---
+title: "GCP Deployment & 30s Load Balancer Timeout"
+aliases: [gcp, gcp-timeout, load-balancer-timeout, websocket-polling]
+tags: [gcp, deployment, websocket, polling, timeout, architecture]
+sources: [01 Projects/modcomms, 01 Projects/semblance]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# GCP Deployment & 30s Load Balancer Timeout
+
+GCP's HTTP(S) Load Balancer drops connections after 30 seconds. Long-running AI tasks that relied on WebSocket streaming were breaking in production.
+
+## Key Takeaways
+- GCP HTTP(S) LB has a **hard 30-second timeout** on backend connections
+- WebSocket connections through GCP LB are killed at 30s → client sees disconnect
+- Fix: replace WebSocket delivery with **HTTP polling** (client polls `/status` endpoint)
+- This affects both Mod Comms and Semblance — same root cause, same fix
+- Task results must be stored server-side (DB or Redis) so the client can retrieve them on poll
+
+## The Problem
+```
+Client → WebSocket → GCP LB → FastAPI/Quart backend
+                              [30 seconds pass]
+           GCP LB KILLS CONNECTION ← AI task not done yet
+Client sees error, no result delivered
+```
+
+## The Fix: HTTP Polling
+```
+1. Client POSTs task → backend returns { task_id }
+2. Client polls GET /tasks/{task_id}/status every 2s
+3. Backend stores result in DB/Redis when task completes
+4. On 'complete' status, client fetches full result
+```
+
+```python
+# FastAPI endpoint pattern
+@router.post("/analyze")
+async def start_analysis(proof: UploadFile):
+    task_id = str(uuid4())
+    background_tasks.add_task(run_ai_analysis, task_id, proof)
+    return {"task_id": task_id}
+
+@router.get("/tasks/{task_id}/status")
+async def get_status(task_id: str):
+    result = await db.get_task(task_id)
+    return {"status": result.status, "data": result.data if result.done else None}
+```
+
+```js
+// Frontend polling
+async function pollTask(taskId) {
+  const { status, data } = await api.get(`/tasks/${taskId}/status`)
+  if (status === 'complete') return data
+  await sleep(2000)
+  return pollTask(taskId)
+}
+```
+
+## Projects Affected
+- [[01 Projects/modcomms/Mod Comms|Mod Comms]] — switched WebSocket → REST polling (2026-03-18, critical fix)
+- [[01 Projects/semblance/Semblance|Semblance]] — migrated all async LLM routes (2026-03-23, critical fix)
+
+## Server Details
+- Both Mod Comms and Semblance deploy to GCP
+- Sandbox NotebookLM and Cinema Studio Pro are on `ai-sandbox.oliver.solutions` (optical-web-1) — different infrastructure, different rules
+
+## Secondary GCP Notes
+- WebSocket keepalive heartbeat should be ≤25s (Mod Comms tuned to 10s)
+- Consider Cloud Run for serverless deployments — has configurable timeout up to 3600s
+- GCP LB timeout is configurable in Backend Service settings, but changing it requires infra access
+
+## Gotchas & Lessons
+- Increasing WebSocket heartbeat (25s → 10s) was not enough — had to drop WebSocket entirely
+- Don't assume WebSocket works through GCP LB without testing with real LB (not just local)
+- Any future GCP project with AI tasks >5s should default to HTTP polling architecture
+- Socket.IO fallback (polling mode) is NOT the same as HTTP polling for results — Socket.IO still goes through LB
+
+## Related
+- [[wiki/architecture/multi-agent-ai-systems|multi-agent-ai-systems]] — the pattern that triggers long tasks
+- [[wiki/tech-patterns/redis-celery-worker-queue|redis-celery-worker-queue]] — storing results for polling
+- [[wiki/tech-patterns/python-ai-agents|python-ai-agents]] — the long-running AI calls
diff --git a/wiki/architecture/hotfolder-daemon.md b/wiki/architecture/hotfolder-daemon.md
new file mode 100644
index 0000000..a036066
--- /dev/null
+++ b/wiki/architecture/hotfolder-daemon.md
@@ -0,0 +1,97 @@
+---
+title: "Hotfolder Daemon — Box Monitoring"
+aliases: [hotfolder, daemon, box-monitor, systemd-daemon]
+tags: [hotfolder, daemon, systemd, box, ford, automation, python]
+sources: [01 Projects/ford_qc, 01 Projects/ford-gechub-sftp]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# Hotfolder Daemon — Box Monitoring
+
+Persistent background daemon that watches a Box folder for new files, processes them automatically, and archives the originals.
+
+## Key Takeaways
+- Ford QC runs as a **systemd service** on a production server — survives reboots
+- Hotfolder pattern: poll Box folder → detect new files → process → upload result → archive original
+- Archive processed files immediately to prevent reprocessing
+- CLI mode (single file) and daemon mode (continuous) — same core logic, different entry points
+- Ford SFTP uses daemon mode (`python main.py --daemon`) without systemd (simpler)
+
+## When to Use
+Any automation that must respond to files dropped in a Box folder without human intervention.
+
+## Architecture
+
+### Ford QC Hotfolder
+```
+Box folder 332861865120 (hotfolder, poll every N seconds)
+    ↓ detect new ZIP
+ford_qc_box_hotfolder_process.py
+    ↓
+qc_engine.py → 13 QC check modules (checks/)
+    ↓
+JSON results → HTML report (html_reporter.py)
+    ↓
+Upload HTML report → Box 332864939558 (reports)
+Archive ZIP → Box 332861653811 (archive)
+```
+
+### Ford SFTP Daemon
+```
+Box folders (PROD / EDU / QA — 3 separate targets)
+    ↓ poll for new ZIPs
+ford_qc_box_hotfolder_process.py
+    ↓
+Download ZIP → Upload to GECHUB SFTP (paramiko)
+    ↓
+Archive in Box
+    ↓
+Email notification (Mailgun)
+```
+
+## Systemd Service (Ford QC)
+```bash
+# /etc/systemd/system/ford-qc-hotfolder.service
+[Unit]
+Description=Ford QC Hotfolder Daemon
+
+[Service]
+ExecStart=/usr/bin/python3 /path/to/ford_qc_box_hotfolder_process.py
+Restart=always
+
+[Install]
+WantedBy=multi-user.target
+
+# Commands
+sudo systemctl start ford-qc-hotfolder.service
+sudo systemctl enable ford-qc-hotfolder.service   # start on boot
+sudo systemctl status ford-qc-hotfolder.service
+```
+
+## Projects Using This Pattern
+- [[01 Projects/ford_qc/Ford QC System|Ford QC System]] — systemd service, 13 QC check modules, Box JWT auth, HTML reports
+- [[01 Projects/ford-gechub-sftp/Ford SFTP Transfer|Ford SFTP Transfer]] — `--daemon` flag, 3 environments, SFTP upload, Mailgun notifications
+
+## QC Module Pattern (Ford QC)
+```python
+# qc_module.py
+class QCModule:
+    def check(self, zip_path: str, profile: dict) -> dict:
+        """Returns {"passed": bool, "issues": [...]}"""
+
+# 13 modules in checks/:
+# image resolution, format, file size, layer depth, colour existence,
+# linking validation, MEC/BAU compliance, powertrain validation, lifestyle inventory
+```
+
+## Gotchas & Lessons
+- Always archive processed files — the daemon will reprocess on next poll if not archived
+- Box JWT config (`ford_box_config.json`) must not be committed to git
+- Ranger ptvl pattern is configurable via profile JSON — don't hardcode (fix: 2026-03-16)
+- Three environments (PROD/EDU/QA) have completely separate Box folder IDs and SFTP targets
+- Mailgun integration: keep email templates simple — Mailgun rejects complex HTML occasionally
+
+## Related
+- [[wiki/tech-patterns/box-api-integration|box-api-integration]] — Box auth and folder IDs
+- [[wiki/client-knowledge/ford|ford]] — Ford-specific context and environment details
diff --git a/wiki/architecture/multi-agent-ai-systems.md b/wiki/architecture/multi-agent-ai-systems.md
new file mode 100644
index 0000000..f2d2716
--- /dev/null
+++ b/wiki/architecture/multi-agent-ai-systems.md
@@ -0,0 +1,86 @@
+---
+title: "Multi-Agent AI Systems"
+aliases: [multi-agent, ai-agents, parallel-agents]
+tags: [ai, multi-agent, architecture, gemini, gpt]
+sources: [01 Projects/modcomms, 01 Projects/semblance, 01 Projects/enterprise-ai-hub-nexus]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# Multi-Agent AI Systems
+
+Pattern where multiple specialized AI agents run in parallel, with a lead/orchestrator agent synthesizing results.
+
+## Key Takeaways
+- Parallel specialist agents + lead synthesizer = reliable, multi-perspective analysis
+- HTTP polling (not WebSocket) for delivering async agent results on GCP
+- Each agent should have a focused, single-concern prompt (Legal, Brand, Tone, Channel)
+- Autonomous orchestration needs explicit "next speaker" logic to prevent infinite loops
+- Background task execution (`ai_runner_service.py`) keeps agents non-blocking
+
+## When to Use
+- Content review requiring multiple compliance dimensions (legal + brand + tone)
+- Focus group simulation (multi-persona conversations)
+- Any task benefiting from multiple independent perspectives before synthesis
+
+## Architecture Patterns
+
+### Pattern 1: Parallel Specialists + Lead (Mod Comms)
+```
+Input (proof image/PDF)
+    ↓
+┌──────────────────────────────────┐
+│ Agent 1: Legal compliance        │
+│ Agent 2: Brand adherence         │  ← run in PARALLEL
+│ Agent 3: Tone of voice           │
+│ Agent 4: Channel suitability     │
+└──────────────────────────────────┘
+    ↓
+Lead Agent: synthesize verdict
+    ↓
+Result (via HTTP polling)
+```
+
+### Pattern 2: Autonomous Multi-Persona (Semblance)
+```
+Input (discussion brief)
+    ↓
+Persona generator (Gemini) → N personas
+    ↓
+Conversation controller
+    ├── conversation_decision_service.py  ← next speaker logic
+    ├── conversation_context_service.py   ← shared state + history
+    └── ai_runner_service.py              ← background task execution
+    ↓
+Socket.IO → frontend (real-time)
+    ↓
+Theme extraction + analytics
+```
+
+### Pattern 3: RAG + Structured AI (Enterprise Nexus)
+```
+Document upload → Firecrawl crawl
+    ↓
+AI content structuring (pre-indexing)
+    ↓
+Qdrant vector DB (10-page batch merge)
+    ↓
+Query → vector search → LLM synthesis
+```
+
+## Projects Using This Pattern
+- [[01 Projects/modcomms/Mod Comms|Mod Comms]] — 4 parallel specialist agents + lead synthesizer (Gemini Pro/Flash)
+- [[01 Projects/semblance/Semblance|Semblance]] — Autonomous focus group with N personas (Gemini 3 Pro, GPT-4.1, GPT-5.2)
+- [[01 Projects/enterprise-ai-hub-nexus/Enterprise AI Hub Nexus|Enterprise Nexus]] — RAG with structured AI pre-processing (Qdrant)
+
+## Gotchas & Lessons
+- GCP 30s LB timeout kills streaming delivery — always use HTTP polling for agent results (see [[wiki/architecture/gcp-deployment-lb-timeout|gcp-deployment-lb-timeout]])
+- Semblance: naive vs aware datetime crash — always use timezone-aware datetimes in async contexts
+- "Next speaker" logic in autonomous mode must have termination conditions to prevent infinite loops
+- Cross-loop WebSocket emit in Semblance was unreliable — polling fallback was more stable
+- Orphaned vectors in Qdrant need periodic cleanup (Enterprise Nexus has a "Purge orphaned vectors" button)
+
+## Related
+- [[wiki/architecture/gcp-deployment-lb-timeout|gcp-deployment-lb-timeout]] — critical deployment constraint
+- [[wiki/architecture/rag-architecture|rag-architecture]] — retrieval-augmented generation
+- [[wiki/tech-patterns/python-ai-agents|python-ai-agents]] — model selection + structured output
diff --git a/wiki/architecture/rag-architecture.md b/wiki/architecture/rag-architecture.md
new file mode 100644
index 0000000..a362c0a
--- /dev/null
+++ b/wiki/architecture/rag-architecture.md
@@ -0,0 +1,100 @@
+---
+title: "RAG Architecture (Retrieval-Augmented Generation)"
+aliases: [rag, vector-search, knowledge-retrieval, qdrant]
+tags: [rag, qdrant, vector-search, llamaindex, firecrawl, architecture]
+sources: [01 Projects/enterprise-ai-hub-nexus, 01 Projects/sandbox-notebookllamalm-nextjs, 01 Projects/Oliver-ai-bot_2.0]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# RAG Architecture
+
+Knowledge retrieval pipeline: ingest documents → chunk + embed → store in vector DB → retrieve relevant chunks → LLM synthesis.
+
+## Key Takeaways
+- Qdrant is the vector DB of choice for Oliver RAG projects
+- AI content structuring **before** indexing improves retrieval quality (Enterprise Nexus)
+- Firecrawl `/v1/crawl` handles recursive website crawling for knowledge ingestion
+- Batch processing (10 pages at a time) for document merging avoids context overflow
+- Orphaned vectors accumulate over time — need periodic cleanup
+- LlamaIndex abstracts multi-model RAG (Sandbox NotebookLM)
+
+## When to Use
+When users need to query a large body of documents in natural language.
+
+## Architecture
+
+### Enterprise Nexus RAG Pipeline
+```
+Sources:
+  - File upload (PDF, Word, etc.)
+  - SharePoint sync (Celery beat)
+  - Website crawl (Firecrawl /v1/crawl)
+    ↓
+AI content structuring (pre-processing pass)
+    ↓
+Qdrant (vector store)
+    ↓ on query:
+Vector similarity search → top-k chunks
+    ↓
+LLM synthesis (prompt + chunks)
+    ↓
+Response to user
+```
+
+### Sandbox NotebookLM RAG
+```
+Document upload → LlamaIndex ingestion
+    ↓
+llm_factory.get_llm_by_type() → multi-model
+    ↓
+7 studio generators (flashcards, quiz, mindmap, slides, report, infographic, datatable)
+    ↓
+ElevenLabs (podcast audio synthesis)
+```
+
+## Key Components
+
+### Qdrant Operations
+```python
+# Upsert vectors
+client.upsert(collection_name="knowledge", points=[
+    PointStruct(id=doc_id, vector=embedding, payload={"text": chunk, "source": filename})
+])
+
+# Search
+hits = client.search(collection_name="knowledge", query_vector=query_embedding, limit=5)
+chunks = [hit.payload["text"] for hit in hits]
+
+# Cleanup orphaned vectors
+client.delete(collection_name="knowledge", points_selector=orphaned_ids)
+```
+
+### Firecrawl Site Crawl
+```python
+response = firecrawl.crawl_url(url, params={"crawlOptions": {"maxDepth": 3}})
+# Returns list of {url, markdown, metadata}
+```
+
+### AI Pre-structuring (Enterprise Nexus)
+- Run LLM pass on raw document before chunking
+- Extracts structured sections, improves chunk boundaries
+- Batch: 10 pages at a time to stay within context window
+
+## Projects Using This Pattern
+- [[01 Projects/enterprise-ai-hub-nexus/Enterprise AI Hub Nexus|Enterprise AI Hub Nexus]] — Qdrant + Firecrawl + AI pre-structuring + SharePoint sync + Celery
+- [[01 Projects/sandbox-notebookllamalm-nextjs/Sandbox NotebookLM|Sandbox NotebookLM]] — LlamaIndex multi-model + 7 studio generators + ElevenLabs podcast
+- [[01 Projects/Oliver-ai-bot_2.0/Oliver AI Bot 2.0|Oliver AI Bot 2.0]] — RAG mode (85% complete)
+
+## Gotchas & Lessons
+- Qdrant vectors are NOT automatically deleted when source documents are removed — implement orphaned vector cleanup
+- AI pre-structuring significantly improves RAG quality but adds ~2-3s per document at ingest
+- SharePoint sync token refresh must be proactive — Celery beat jobs run after tokens may have expired
+- M365 delegated scopes + `offline_access` required for refresh tokens (Enterprise Nexus)
+- `max_tokens` 8192 was too low for from-template PPTX generation in Sandbox — increased to 16000
+
+## Related
+- [[wiki/tech-patterns/redis-celery-worker-queue|redis-celery-worker-queue]] — SharePoint sync scheduling
+- [[wiki/tech-patterns/azure-ad-msal-auth|azure-ad-msal-auth]] — M365 auth for SharePoint
+- [[wiki/tech-patterns/python-ai-agents|python-ai-agents]] — LLM integration details
+- [[wiki/architecture/multi-agent-ai-systems|multi-agent-ai-systems]] — related AI pattern
diff --git a/wiki/client-knowledge/_index.md b/wiki/client-knowledge/_index.md
new file mode 100644
index 0000000..871101f
--- /dev/null
+++ b/wiki/client-knowledge/_index.md
@@ -0,0 +1,30 @@
+---
+title: "Client Knowledge Index"
+description: "What Vadym knows about working with specific clients — tech preferences, quirks, and constraints"
+tags: [index, client-knowledge]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# Client Knowledge
+
+Per-client notes for clients with 2+ active projects. Covers tech preferences, constraints, and lessons learned.
+
+## Clients
+
+| Client | Projects | Key Tech | Article |
+|--------|---------|---------|---------|
+| Ford | Ford QC, Ford SFTP | Box API, SFTP, systemd, Python | [[wiki/client-knowledge/ford\|ford]] |
+| H&M | O2E Tool, EMS Report | One2Edit API, Python, JSON | [[wiki/client-knowledge/hm\|hm]] |
+| L'Oréal | Global Kickoff, SLA Calculator | Box API, PHP, Make.com, Docker | [[wiki/client-knowledge/loreal\|loreal]] |
+
+## Single-Project Clients
+These clients have only one project — context lives in the project note:
+- **3M** → [[01 Projects/3m-portal/3M OMG Portal|3M OMG Portal]] — One2Edit proxy
+- **Barclays** → [[01 Projects/modcomms/Mod Comms|Mod Comms]] — AI proof review, GCP
+- **BAIC** → [[01 Projects/baic_dashboard/BAIC Dashboard|BAIC Dashboard]] — Make.com + Azure AD
+- **PIMCO** → [[01 Projects/pimco-charts/PIMCO Chart Generator|PIMCO Charts]] — SVG chart generation
+- **Ferrero** → [[01 Projects/ferrero-ac-creator/Ferrero AC Booking|Ferrero AC Booking]] — Box API + CSV
+- **Solventum** → [[01 Projects/solventum-image-metadata/Solventum Image Metadata|Solventum Metadata]] — OpenAI + enterprise
+- **LUSA** → [[01 Projects/lusa-back-planner/LUSA Back Planner|LUSA Back Planner]] — Timeline + PDF
+- **Dow Jones / WSJ** → [[01 Projects/wsj-filenaming/WSJ File Naming Tool|WSJ File Naming]] — Gemini AI naming
diff --git a/wiki/client-knowledge/ford.md b/wiki/client-knowledge/ford.md
new file mode 100644
index 0000000..247e5b7
--- /dev/null
+++ b/wiki/client-knowledge/ford.md
@@ -0,0 +1,66 @@
+---
+title: "Ford — Client Knowledge"
+aliases: [ford, ford-motor]
+tags: [client, ford, box, sftp, qc, automation]
+sources: [01 Projects/ford_qc, 01 Projects/ford-gechub-sftp]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# Ford
+
+Oliver Agency works with Ford on asset pack automation — QC validation and delivery to Ford's GECHUB SFTP system.
+
+## Key Takeaways
+- Ford has **3 environments**: PROD, EDU, QA — each has separate Box folders and SFTP targets
+- Asset packs are ZIP files — QC runs 13 checks before delivery
+- Ford uses GECHUB as their SFTP-based delivery system
+- Box JWT service account (`ford_box_config.json`) — never commit this file
+- The Ford QC daemon runs as a **systemd service** on the production server
+
+## Projects
+| Project | Purpose | Entry Point |
+|---------|---------|-------------|
+| [[01 Projects/ford_qc/Ford QC System\|Ford QC System]] | QC validation of asset pack ZIPs | systemd or `python qc_engine.py` |
+| [[01 Projects/ford-gechub-sftp/Ford SFTP Transfer\|Ford SFTP Transfer]] | Box → GECHUB SFTP transfer | `python main.py [--daemon]` |
+
+## Environment Details
+| Env | Purpose | Box Folder | SFTP Target |
+|-----|---------|-----------|-------------|
+| PROD | Production assets | Box/PROD | GECHUB PROD |
+| EDU | Educational assets | Box/EDU | GECHUB EDU |
+| QA | QA testing | Box/QA | GECHUB QA |
+
+## Box Folder IDs (Ford QC)
+| Folder | ID |
+|--------|-----|
+| Input hotfolder | 332861865120 |
+| Reports output | 332864939558 |
+| Archive | 332861653811 |
+
+## Ford QC Checks (13 modules)
+- Image resolution, format, file size
+- Layer depth, colour existence
+- Linking validation
+- MEC/BAU compliance
+- Powertrain validation
+- Lifestyle inventory
+- HTML report generation
+
+## Tech Preferences
+- Box API (JWT auth) — file exchange platform
+- SFTP (GECHUB) — delivery system
+- systemd for production daemons
+- Email notifications via Mailgun
+- QC profiles in JSON (configurable per asset type)
+- HTML reports (not PDFs)
+
+## Quirks & Lessons
+- Ranger ptvl pattern must be configurable via QC profile — don't hardcode it (fix: 2026-03-16)
+- Always archive processed ZIPs immediately — daemon reprocesses if not archived
+- Ford has 3 completely separate environments — changes in PROD profiles don't affect EDU/QA
+- GECHUB SFTP credentials differ per environment
+
+## Related
+- [[wiki/architecture/hotfolder-daemon|hotfolder-daemon]] — the daemon pattern
+- [[wiki/tech-patterns/box-api-integration|box-api-integration]] — Box details
diff --git a/wiki/client-knowledge/hm.md b/wiki/client-knowledge/hm.md
new file mode 100644
index 0000000..b0f30c4
--- /dev/null
+++ b/wiki/client-knowledge/hm.md
@@ -0,0 +1,53 @@
+---
+title: "H&M — Client Knowledge"
+aliases: [hm, h&m, handm]
+tags: [client, hm, one2edit, ems, product-review]
+sources: [01 Projects/hm-o2e-tool, 01 Projects/hm_ems_report]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# H&M
+
+Oliver Agency works with H&M on document translation/relinking (One2Edit platform) and product data review workflows.
+
+## Key Takeaways
+- H&M uses **One2Edit** as their document management/translation platform
+- EMS product review is multi-language (up to 4 languages simultaneously)
+- Both H&M tools are lightweight — no Docker, minimal infrastructure
+- Product data lives in a master JSON structure; edits save back to master
+
+## Projects
+| Project | Purpose | Entry Point |
+|---------|---------|-------------|
+| [[01 Projects/hm-o2e-tool/HM O2E Tool\|H&M O2E Tool]] | Image relinking + document export via One2Edit | Static / `python -m http.server` |
+| [[01 Projects/hm_ems_report/HM EMS Report Tool\|H&M EMS Report Tool]] | Multi-language product review and approval | `python server.py` |
+
+## H&M O2E Tool Capabilities
+- Automated image relinking (by File or by Folder)
+- Document export from One2Edit
+- Image status checking
+- Static — no backend required
+
+## H&M EMS Review Tool Capabilities
+- Side-by-side review of product names + prices in up to 4 languages
+- Inline editing
+- Approval workflow → saves to `Master_Json/`
+- Full change log
+- Campaign images with enlarge preview
+
+## Tech Preferences
+- One2Edit API (`oliver.one2edit.com/v3/Api.php`)
+- Lightweight tools — no Docker, no complex build
+- JSON as data store (not a database)
+- Python simple HTTP server for local serving
+
+## Quirks & Lessons
+- One2Edit is shared with 3M — same platform, different use cases
+- H&M O2E tool can run fully static (open `index.html`) for most operations
+- EMS tool: master JSON is the source of truth — backup before any bulk operations
+- Multi-language display: columns for each language, all visible simultaneously
+
+## Related
+- [[wiki/tech-patterns/one2edit-api|one2edit-api]] — One2Edit platform details
+- [[wiki/client-knowledge/ford|ford]] — another client using Box-based workflows
diff --git a/wiki/client-knowledge/loreal.md b/wiki/client-knowledge/loreal.md
new file mode 100644
index 0000000..1213304
--- /dev/null
+++ b/wiki/client-knowledge/loreal.md
@@ -0,0 +1,56 @@
+---
+title: "L'Oréal — Client Knowledge"
+aliases: [loreal, l'oreal, loreal-global]
+tags: [client, loreal, box, make, php, sla, global-to-local]
+sources: [01 Projects/loreal-global-kickoff, 01 Projects/loreal-sla-calculator]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# L'Oréal
+
+Oliver Agency works with L'Oréal on campaign asset management (global → local transformation) and SLA timeline tooling.
+
+## Key Takeaways
+- L'Oréal uses **Box** for campaign asset exchange and folder hierarchy encodes campaign numbers
+- **Make.com** is used as the automation backbone (webhook on submission)
+- The "Global to Local" workflow transforms global campaign CSVs into regional market CSVs
+- SLA calculator replaces Excel — 9 brief types, 8 stages, Gantt view
+- PHP backend for the global kickoff tool (not Python/Node)
+
+## Projects
+| Project | Purpose | Entry Point |
+|---------|---------|-------------|
+| [[01 Projects/loreal-global-kickoff/Loreal Global Kickoff\|L'Oréal OMG Assistant Global]] | Asset submission + global-to-local CSV transform | PHP server |
+| [[01 Projects/loreal-sla-calculator/Loreal SLA Calculator\|L'Oréal SLA Calculator]] | SLA timeline calculator for eCom Content Factory PMs | `docker compose up` |
+
+## L'Oréal OMG Assistant Capabilities
+- **Master Global Asset Submission** — submit Box assets with metadata
+- Auto-populate campaign number from Box folder hierarchy
+- **Global to Local** — transform global CSVs into regional market CSVs
+- Webhook to Make.com on submission
+
+## L'Oréal SLA Calculator Capabilities
+- Replaces Excel for SLA timeline management
+- 9 brief types × 8 stages
+- Gantt chart view + date verdict
+- CSV + iCal export
+- Client Estimator (`client.html`) — simplified 4-input version
+
+## Tech Preferences
+- PHP for server-rendered web apps (not Python or Node)
+- Box API for file storage and folder-based campaign organization
+- Make.com for automation/workflow orchestration
+- CSV as data exchange format for global → local transforms
+- Docker for SLA calculator (simpler tool)
+
+## Quirks & Lessons
+- Campaign number is encoded in the Box folder hierarchy — must traverse parent folders to extract it
+- Make.com webhook triggers on every submission — design idempotent handlers
+- The "Global to Local" transform needs to understand regional market codes in CSV headers
+- SLA calculator has two UX modes: full PM calculator vs simplified client estimator
+- L'Oréal uses eCom Content Factory workflow — their PM team are the primary users of the SLA tool
+
+## Related
+- [[wiki/tech-patterns/box-api-integration|box-api-integration]] — Box API patterns
+- [[wiki/client-knowledge/ford|ford]] — another Box-heavy client
diff --git a/wiki/projects-overview/_index.md b/wiki/projects-overview/_index.md
new file mode 100644
index 0000000..6b70feb
--- /dev/null
+++ b/wiki/projects-overview/_index.md
@@ -0,0 +1,79 @@
+---
+title: "Projects Overview"
+description: "All Oliver Agency projects — navigable map with tech stack and status"
+tags: [projects, oliver, index]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# Projects Overview
+
+All active projects at `/Volumes/SSD/Projects/Oliver/`. One row per project.
+
+## Key Takeaways
+- 36 projects total; majority use Docker Compose as deploy mechanism
+- FastAPI + React/Vite + Docker is the dominant full-stack pattern
+- Azure AD / MSAL appears in 6+ projects (Oliver internal tools)
+- GCP is the primary production server for external-facing tools
+- Box API used by Ford, Ferrero, L'Oréal integrations
+
+## Project Map
+
+| Project | Client | Status | Tech Stack | Server / Run |
+|---------|--------|--------|-----------|--------------|
+| [[01 Projects/3m-portal/3M OMG Portal\|3M OMG Portal]] | 3M | Active | Node.js, Vanilla JS, One2Edit API | `node server.js` :3000 |
+| [[01 Projects/apac-ops-bot/APAC Ops Bot\|APAC Ops Bot]] | Oliver | Active | TBD | TBD |
+| [[01 Projects/DevOps_Click_UP_sync/DevOps ClickUp Sync\|DevOps ClickUp Sync]] | Oliver | Active | Python, FastAPI, SQLite, Tailwind, Docker | `docker compose up` :8080 |
+| [[01 Projects/Oliver-ai-bot_2.0/Oliver AI Bot 2.0\|Oliver AI Bot 2.0]] | Oliver | Active | Next.js, FastAPI, Python, Docker | `docker compose up` |
+| [[01 Projects/ac-helper/AC Helper\|AC Helper]] | Oliver | Active | Python, Gemini, HTML/JS, Tailwind, Docker | `docker compose up` |
+| [[01 Projects/ac-tool/AC Tool\|AC Tool]] | Oliver | Active | Docker | `docker compose up` |
+| [[01 Projects/baic_dashboard/BAIC Dashboard\|BAIC Dashboard]] | BAIC | Active | React, Vite, Flask, Recharts, Azure AD, Make.com | `python app.py` + `npm run dev` |
+| [[01 Projects/build-a-squad/Build A Squad\|Build A Squad]] | Oliver | Active | React 19, TypeScript, Vite, Gemini | `npm run dev` |
+| [[01 Projects/cc-dashboard/CC Dashboard\|CC Dashboard]] | Oliver | Active | Docker | `docker compose up` |
+| [[01 Projects/cinema-studio-pro/Cinema Studio Pro\|Cinema Studio Pro (Lux Studio)]] | Oliver | Active | React, Vite, PHP, Azure AD, Imagen 3, Veo, Kling | `./setup.sh` / FileZilla+SSH |
+| [[01 Projects/enterprise-ai-hub-nexus/Enterprise AI Hub Nexus\|Enterprise AI Hub Nexus]] | Oliver | **Production** | Next.js 14, FastAPI, PostgreSQL, Redis, Qdrant, Azure AD, Celery, Firecrawl | `docker compose up` |
+| [[01 Projects/ferrero-ac-creator/Ferrero AC Booking Tool\|Ferrero AC Booking Tool]] | Ferrero | Active | Node.js, HTML/JS, Box API | `node server.js` :3456 |
+| [[01 Projects/ford-gechub-sftp/Ford SFTP Transfer\|Ford Asset Pack SFTP Transfer]] | Ford | Active | Python, Box API, SFTP, Mailgun | `python main.py [--daemon]` |
+| [[01 Projects/ford_qc/Ford QC System\|Ford QC System]] | Ford | Active | Python, Box API, systemd | `systemctl start ford-qc-hotfolder` |
+| [[01 Projects/gmal-scope-builder/GMAL Scope Builder\|GMAL Scope Builder]] | Oliver | Active | React, FastAPI, PostgreSQL, Claude Opus 4.6, Azure SSO, Docker | `docker compose up` :3010/:8001 |
+| [[01 Projects/hm-o2e-tool/HM O2E Tool\|H&M O2E Tool]] | H&M | Active | HTML/JS, One2Edit API | Static / `python -m http.server` |
+| [[01 Projects/hm_ems_report/HM EMS Report Tool\|H&M EMS Report Tool]] | H&M | Active | Python, HTML/JS, JSON | `python server.py` |
+| [[01 Projects/homepage/Homepage Dashboard\|Homepage Dashboard]] | Oliver | Active | Node.js, Docker, YAML | `docker compose up` |
+| [[01 Projects/loreal-global-kickoff/Loreal Global Kickoff\|L'Oréal OMG Assistant Global]] | L'Oréal | Active | PHP, Box API, Make.com | PHP server |
+| [[01 Projects/loreal-sla-calculator/Loreal SLA Calculator\|L'Oréal SLA Calculator]] | L'Oréal | Active | HTML/JS, Docker, Node.js | `docker compose up` |
+| [[01 Projects/lusa-back-planner/LUSA Back Planner\|LUSA Back Planner]] | LUSA | Active | React 19, TypeScript, Vite, Tailwind, shadcn/ui, pdfjs, jsPDF | `npm run dev` |
+| [[01 Projects/md-to-word/MD to Word\|MD to Word Converter]] | Oliver | Active | Python | `python convert.py` |
+| [[01 Projects/modcomms/Mod Comms\|Mod Comms]] | Barclays | Active | React, FastAPI, PostgreSQL, Gemini, Azure AD, Docker | `docker compose up` — GCP |
+| [[01 Projects/olivas/OliVAS\|OliVAS]] | Oliver | Active | Python, FastAPI, Claude API, DeepGaze, Docker | `docker compose up` |
+| [[01 Projects/oliver-ai-assistant/Oliver AI Assistant\|Oliver AI Assistant]] | Oliver | Active | TBD | TBD |
+| [[01 Projects/pdf-accessibility/PDF Accessibility Checker\|PDF Accessibility Checker]] | Oliver | Active | Python, PHP, PostgreSQL, Redis, Claude, GCV, Docker | `docker compose up` |
+| [[01 Projects/pimco-charts/PIMCO Chart Generator\|PIMCO Chart Generator]] | PIMCO | Active | Python, FastAPI, Claude API, SVG, Docker | `docker compose up` |
+| [[01 Projects/ppt-tool/Oliver DeckForge\|Oliver DeckForge]] | Oliver | Active | Next.js, FastAPI, nginx, Docker | `docker compose up` |
+| [[01 Projects/presenton/Presenton\|Presenton]] | Oliver | Active | Node.js, Docker | `docker compose up` |
+| [[01 Projects/sandbox-notebookllamalm-nextjs/Sandbox NotebookLM\|Sandbox NotebookLM]] | Oliver | Active | Next.js 15, FastAPI, PostgreSQL, Redis, ElevenLabs, Docker | `docker compose up` — optical-web-1 |
+| [[01 Projects/semblance/Semblance\|Semblance]] | Oliver | Active | React, Quart, Socket.IO, MongoDB, Gemini, OpenAI, Docker | `docker compose up` — GCP |
+| [[01 Projects/smartcrop26/SmartCrop26\|SmartCrop26]] | Oliver | Active | React, TypeScript, Vite, Lovable | `npm run dev` |
+| [[01 Projects/solventum-image-metadata/Solventum Image Metadata\|Solventum Image Metadata Tool]] | Solventum | Active | Python, FastAPI, OpenAI, Docker | `docker compose up` |
+| [[01 Projects/video-accessibility/Video Accessibility Platform\|Video Accessibility Platform]] | Oliver | Active | React, FastAPI, Celery, MongoDB, Redis, Gemini, GCS, Docker | `docker compose up` |
+| [[01 Projects/wsj-filenaming/WSJ File Naming Tool\|WSJ File Naming Tool]] | Dow Jones | Active | Python, HTML/JS, Gemini | `python server.py` |
+
+## Clients
+| Client | Projects | Notes |
+|--------|---------|-------|
+| Oliver Internal | ~25 | Majority of projects |
+| Ford | Ford QC, Ford SFTP | Box API + SFTP automation |
+| H&M | O2E Tool, EMS Report | One2Edit + product review |
+| L'Oréal | Global Kickoff, SLA Calculator | Box API + PHP |
+| Barclays | Mod Comms | GCP deployment |
+| 3M | OMG Portal | One2Edit proxy |
+| BAIC | Dashboard | Make.com + Azure AD |
+| PIMCO | Chart Generator | SVG publication charts |
+| Ferrero | AC Booking | Box API + CSV |
+| Solventum | Image Metadata | OpenAI + enterprise |
+| LUSA | Back Planner | Timeline/PDF tool |
+| Dow Jones / WSJ | File Naming Tool | Gemini AI naming |
+
+## Related
+- [[wiki/tech-patterns/_index|tech-patterns/]] — recurring stack patterns
+- [[wiki/architecture/_index|architecture/]] — cross-cutting architectural decisions
+- [[wiki/client-knowledge/_index|client-knowledge/]] — per-client notes
diff --git a/wiki/tech-patterns/_index.md b/wiki/tech-patterns/_index.md
new file mode 100644
index 0000000..4be6b7a
--- /dev/null
+++ b/wiki/tech-patterns/_index.md
@@ -0,0 +1,36 @@
+---
+title: "Tech Patterns Index"
+description: "Recurring technology stacks across Oliver Agency projects"
+tags: [index, tech-patterns]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# Tech Patterns
+
+Recurring technology stacks used across Oliver Agency projects. Each article covers a specific pattern with projects, gotchas, and code examples.
+
+## Patterns
+
+| Pattern | Description | Projects |
+|---------|-------------|---------|
+| [[wiki/tech-patterns/fastapi-python-docker\|fastapi-python-docker]] | FastAPI + Python + Docker — dominant backend stack | GMAL, Mod Comms, Video Accessibility, OliVAS, PIMCO, Solventum, Enterprise Nexus |
+| [[wiki/tech-patterns/react-vite-typescript\|react-vite-typescript]] | React + Vite + TypeScript — standard Oliver SPA frontend | GMAL, Mod Comms, Semblance, Video Accessibility, LUSA, Build-A-Squad, BAIC |
+| [[wiki/tech-patterns/nextjs-fastapi-fullstack\|nextjs-fastapi-fullstack]] | Next.js + FastAPI — for complex multi-service AI platforms | Enterprise Nexus, Sandbox NotebookLM, Oliver AI Bot 2.0, DeckForge |
+| [[wiki/tech-patterns/azure-ad-msal-auth\|azure-ad-msal-auth]] | Azure AD / MSAL SSO — Oliver internal auth standard | GMAL, Enterprise Nexus, Cinema Studio Pro, Mod Comms, BAIC, Sandbox |
+| [[wiki/tech-patterns/python-ai-agents\|python-ai-agents]] | Claude / Gemini / OpenAI / LlamaIndex integration patterns | GMAL, Mod Comms, Semblance, OliVAS, PIMCO, Video Accessibility, Enterprise Nexus |
+| [[wiki/tech-patterns/redis-celery-worker-queue\|redis-celery-worker-queue]] | Redis + Celery for async long-running tasks | Enterprise Nexus, Video Accessibility, PDF Accessibility |
+| [[wiki/tech-patterns/box-api-integration\|box-api-integration]] | Box API for client asset workflows | Ford QC, Ford SFTP, L'Oréal, Ferrero |
+| [[wiki/tech-patterns/one2edit-api\|one2edit-api]] | One2Edit translation platform API | 3M Portal, H&M O2E Tool |
+| [[wiki/tech-patterns/nodejs-vanilla-proxy\|nodejs-vanilla-proxy]] | Node.js + Vanilla JS lightweight proxy tools | 3M Portal, Ferrero, Homepage |
+
+## Quick Decision Guide
+
+```
+New project → what stack?
+├── Complex AI platform, multi-user → nextjs-fastapi-fullstack
+├── Standard tool with UI → fastapi-python-docker + react-vite-typescript
+├── Simple client portal / proxy → nodejs-vanilla-proxy
+├── Static page, no backend → plain HTML/JS
+└── Needs auth? → azure-ad-msal-auth
+```
diff --git a/wiki/tech-patterns/azure-ad-msal-auth.md b/wiki/tech-patterns/azure-ad-msal-auth.md
new file mode 100644
index 0000000..bfad756
--- /dev/null
+++ b/wiki/tech-patterns/azure-ad-msal-auth.md
@@ -0,0 +1,83 @@
+---
+title: "Azure AD / MSAL Authentication"
+aliases: [azure-ad, msal, sso, pkce]
+tags: [azure-ad, msal, auth, sso, pkce, security]
+sources: [01 Projects/gmal-scope-builder, 01 Projects/enterprise-ai-hub-nexus, 01 Projects/cinema-studio-pro, 01 Projects/modcomms, 01 Projects/baic_dashboard, 01 Projects/sandbox-notebookllamalm-nextjs]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# Azure AD / MSAL Authentication
+
+Oliver Agency standard SSO. MSAL handles token acquisition; backend validates via Azure AD JWT.
+
+## Key Takeaways
+- All Oliver-internal tools requiring login use Azure AD (MSAL)
+- MSAL v5 introduced breaking changes — `clearCache()` for logout, NOT `logout()`
+- Send **ID token** to backend, NOT the Graph access token
+- Use `DEV_AUTH_BYPASS=true` env var for local dev to skip auth
+- Backend validates via `JWTValidator.php` (PHP) or Azure AD middleware (Python)
+- PKCE flow is used for SPAs (Enterprise Nexus); standard OAuth2 for server-side
+
+## When to Use
+Any Oliver internal tool that needs to know who the user is. Skip for pure tools/utilities.
+
+## Key Details
+
+### Frontend Config
+```js
+// authConfig.js / authConfig.ts
+{
+  auth: {
+    clientId: process.env.AZURE_CLIENT_ID,
+    authority: `https://login.microsoftonline.com/${TENANT_ID}`,
+    redirectUri: window.location.origin
+  },
+  cache: { cacheLocation: "sessionStorage" }
+}
+```
+
+### Required Env Vars
+```env
+AZURE_TENANT_ID=
+AZURE_CLIENT_ID=
+AZURE_CLIENT_SECRET=   # server-side only
+DEV_AUTH_BYPASS=true   # local dev only
+```
+
+### Scopes
+- **M365 integration** requires delegated scopes + `offline_access` (Enterprise Nexus)
+- **Graph API** access requires explicit scope grants in Azure App Registration
+
+### MSAL v5 Breaking Changes
+```js
+// Old (MSAL v4):
+instance.logout()
+// New (MSAL v5):
+instance.clearCache()
+```
+
+### Token Validation (Backend)
+- **PHP:** `JWTValidator.php` + `AuthMiddleware.php` (Cinema Studio Pro)
+- **Python:** Azure AD middleware checks Bearer token against tenant
+
+## Projects Using This Pattern
+- [[01 Projects/gmal-scope-builder/GMAL Scope Builder|GMAL Scope Builder]] — MSAL v5, DEV_AUTH_BYPASS, Azure SSO
+- [[01 Projects/enterprise-ai-hub-nexus/Enterprise AI Hub Nexus|Enterprise Nexus]] — PKCE, delegated scopes, offline_access, M365
+- [[01 Projects/cinema-studio-pro/Cinema Studio Pro|Cinema Studio Pro]] — MSAL + JWTValidator.php backend
+- [[01 Projects/modcomms/Mod Comms|Mod Comms]] — Azure AD, DISABLE_AUTH for local dev
+- [[01 Projects/baic_dashboard/BAIC Dashboard|BAIC Dashboard]] — MSAL, Azure App Registration required
+- [[01 Projects/sandbox-notebookllamalm-nextjs/Sandbox NotebookLM|Sandbox NotebookLM]] — Microsoft SSO + signup/login
+
+## Gotchas & Lessons
+- MSAL v5 `clearCache()` not `logout()` — fix was needed in GMAL (2026-03-30)
+- Send ID token to backend, not Graph access token — GMAL bug fix (2026-03-30)
+- `needs_reconsent` flag: if Graph token refresh fails, prompt user to re-login (Enterprise Nexus)
+- Azure App Registration must have the delegated scopes explicitly added for M365 features
+- `offline_access` scope required for refresh token to work (Enterprise Nexus M365)
+- Proactive token refresh needed for long-running Celery jobs (Enterprise Nexus SharePoint sync)
+
+## Related
+- [[wiki/tech-patterns/fastapi-python-docker|fastapi-python-docker]] — backend receiving tokens
+- [[wiki/tech-patterns/react-vite-typescript|react-vite-typescript]] — frontend MSAL integration
+- [[wiki/tech-patterns/nextjs-fastapi-fullstack|nextjs-fastapi-fullstack]] — Next.js PKCE flow
diff --git a/wiki/tech-patterns/box-api-integration.md b/wiki/tech-patterns/box-api-integration.md
new file mode 100644
index 0000000..55261c5
--- /dev/null
+++ b/wiki/tech-patterns/box-api-integration.md
@@ -0,0 +1,73 @@
+---
+title: "Box API Integration"
+aliases: [box-api, box]
+tags: [box, api, integration, ford, loreal, ferrero]
+sources: [01 Projects/ford_qc, 01 Projects/ford-gechub-sftp, 01 Projects/loreal-global-kickoff, 01 Projects/ferrero-ac-creator]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# Box API Integration
+
+Box is used as the file storage and workflow layer for several client integrations. Oliver Agency uses Box for asset handoff with Ford, L'Oréal, and Ferrero.
+
+## Key Takeaways
+- Ford uses Box JWT auth (`ford_box_config.json`) — service account model
+- L'Oréal and Ferrero use Box API for folder-based asset submission workflows
+- Ford QC monitors specific Box folder IDs (hardcoded in profile JSON, not config)
+- Box folder IDs are stable identifiers — use them, don't rely on folder names
+
+## When to Use
+Any client project where files are exchanged via Box. Box is the primary asset handoff platform for Ford, L'Oréal, Ferrero.
+
+## Key Details
+
+### Auth Methods
+| Project | Auth Method | Config |
+|---------|------------|--------|
+| Ford QC | JWT (service account) | `ford_box_config.json` |
+| Ford SFTP | Box API | env vars |
+| L'Oréal | Box API | env vars |
+| Ferrero | Box API | env vars |
+
+### Ford Box Folder IDs
+| Purpose | Folder ID |
+|---------|-----------|
+| Input hotfolder | 332861865120 |
+| Reports output | 332864939558 |
+| Archive (processed) | 332861653811 |
+
+### Common Operations
+```python
+# Download file from Box
+client = Client(auth=JWTAuth.from_settings_file('ford_box_config.json'))
+file_content = client.file(file_id).content()
+
+# Upload to Box
+folder = client.folder(folder_id)
+folder.upload_stream(file_stream, 'filename.zip')
+
+# List folder contents
+items = client.folder(folder_id).get_items()
+```
+
+### Make.com Webhook Integration
+L'Oréal Global Kickoff sends Box submission metadata to Make.com via webhook on form submit.
+
+## Projects Using This Pattern
+- [[01 Projects/ford_qc/Ford QC System|Ford QC System]] — JWT auth, monitors 3 folder IDs, uploads HTML reports, archives processed files
+- [[01 Projects/ford-gechub-sftp/Ford SFTP Transfer|Ford SFTP Transfer]] — Download from Box (3 envs), upload to GECHUB SFTP
+- [[01 Projects/loreal-global-kickoff/Loreal Global Kickoff|L'Oréal Global Kickoff]] — Asset submission form + Make.com webhook
+- [[01 Projects/ferrero-ac-creator/Ferrero AC Booking Tool|Ferrero AC Booking]] — "Send to OMG" saves CSV to Box folder
+
+## Gotchas & Lessons
+- JWT config file (`ford_box_config.json`) must not be committed — it's a service account credential
+- Box folder IDs are numeric strings — store in config, not hardcoded in logic
+- Ford has 3 environments (PROD/EDU/QA) each with separate Box folders and SFTP targets
+- Hotfolder daemon should archive processed files to prevent reprocessing (Ford QC pattern)
+- Box webhooks have retry logic — design idempotent handlers
+
+## Related
+- [[wiki/architecture/hotfolder-daemon|hotfolder-daemon]] — Ford QC Box monitoring pattern
+- [[wiki/client-knowledge/ford|ford]] — Ford-specific context
+- [[wiki/client-knowledge/loreal|loreal]] — L'Oréal-specific context
diff --git a/wiki/tech-patterns/fastapi-python-docker.md b/wiki/tech-patterns/fastapi-python-docker.md
new file mode 100644
index 0000000..878c704
--- /dev/null
+++ b/wiki/tech-patterns/fastapi-python-docker.md
@@ -0,0 +1,73 @@
+---
+title: "FastAPI + Python + Docker Stack"
+aliases: [fastapi-docker, python-api]
+tags: [fastapi, python, docker, backend]
+sources: [01 Projects/gmal-scope-builder, 01 Projects/modcomms, 01 Projects/video-accessibility, 01 Projects/pdf-accessibility, 01 Projects/olivas, 01 Projects/pimco-charts, 01 Projects/solventum-image-metadata]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# FastAPI + Python + Docker Stack
+
+The dominant backend pattern at Oliver — Python FastAPI served via Docker Compose, typically paired with React/Vite frontend.
+
+## Key Takeaways
+- Default backend choice for all new Oliver AI tools
+- Uvicorn is the ASGI server; standard port is 8000 or 8001
+- Always use `asyncpg` + SQLAlchemy async when touching PostgreSQL
+- Docker Compose is the standard deploy — backend + frontend + DB in one `compose.yml`
+- `alembic` for database migrations (PostgreSQL projects)
+- `uv` emerging as the package manager of choice (Sandbox NotebookLM uses Python 3.13 + uv)
+
+## When to Use
+Any new Oliver internal tool needing a REST API backend, especially when AI integration is required.
+
+## Key Details
+- **ASGI server:** Uvicorn (`uvicorn app.main:app --reload --host 0.0.0.0 --port 8000`)
+- **Async DB:** SQLAlchemy async + asyncpg (PostgreSQL) or PyMongo (MongoDB)
+- **Migrations:** Alembic (30+ revisions in Enterprise Nexus)
+- **Package manager:** pip / venv standard; `uv` for newer projects
+- **Port conventions:** Backend 8000 or 8001; Frontend 3000, 3010, or 5173
+
+## Projects Using This Pattern
+- [[01 Projects/gmal-scope-builder/GMAL Scope Builder|GMAL Scope Builder]] — FastAPI :8001, PostgreSQL :5433, React :3010
+- [[01 Projects/modcomms/Mod Comms|Mod Comms]] — FastAPI :8000, PostgreSQL, React :3000, GCP
+- [[01 Projects/video-accessibility/Video Accessibility Platform|Video Accessibility]] — FastAPI + Celery workers, MongoDB, Redis
+- [[01 Projects/pdf-accessibility/PDF Accessibility Checker|PDF Accessibility]] — FastAPI + PHP hybrid, PostgreSQL, Redis
+- [[01 Projects/enterprise-ai-hub-nexus/Enterprise AI Hub Nexus|Enterprise Nexus]] — FastAPI + Celery + Qdrant + PostgreSQL + Redis
+- [[01 Projects/olivas/OliVAS|OliVAS]] — FastAPI + DeepGaze + Claude
+- [[01 Projects/pimco-charts/PIMCO Chart Generator|PIMCO Charts]] — FastAPI + Claude API
+- [[01 Projects/solventum-image-metadata/Solventum Image Metadata|Solventum]] — FastAPI + OpenAI
+- [[01 Projects/Oliver-ai-bot_2.0/Oliver AI Bot 2.0|Oliver AI Bot 2.0]] — FastAPI + Next.js
+
+## Standard Docker Compose Structure
+```yaml
+services:
+  backend:
+    build: ./backend
+    ports: ["8000:8000"]
+    env_file: .env
+    depends_on: [db, redis]
+  frontend:
+    build: ./frontend
+    ports: ["3000:3000"]
+  db:
+    image: postgres:16
+    ports: ["5432:5432"]
+  redis:
+    image: redis:7
+```
+
+## Gotchas & Lessons
+- GCP Load Balancer has a 30s timeout — long AI calls must use **HTTP polling**, not WebSocket streaming (learned from Mod Comms + Semblance)
+- Vite proxy timeout must be increased for long AI operations: `proxy timeout: 300000` (5 min) — learned from GMAL
+- `DEV_AUTH_BYPASS` env var pattern for skipping Azure AD auth locally
+- For PostgreSQL, always pin port to non-default (e.g., 5433) to avoid conflicts with local installs
+- When using Alembic: run `alembic upgrade head` inside the container after first start
+
+## Related
+- [[wiki/tech-patterns/react-vite-typescript|react-vite-typescript]] — paired frontend
+- [[wiki/tech-patterns/azure-ad-msal-auth|azure-ad-msal-auth]] — auth layer
+- [[wiki/tech-patterns/redis-celery-worker-queue|redis-celery-worker-queue]] — async task processing
+- [[wiki/architecture/docker-compose-fullstack|docker-compose-fullstack]] — deployment pattern
+- [[wiki/architecture/gcp-deployment-lb-timeout|gcp-deployment-lb-timeout]] — GCP gotcha
diff --git a/wiki/tech-patterns/nextjs-fastapi-fullstack.md b/wiki/tech-patterns/nextjs-fastapi-fullstack.md
new file mode 100644
index 0000000..e9cc23c
--- /dev/null
+++ b/wiki/tech-patterns/nextjs-fastapi-fullstack.md
@@ -0,0 +1,63 @@
+---
+title: "Next.js + FastAPI Fullstack"
+aliases: [nextjs-fastapi, nextjs-stack]
+tags: [nextjs, fastapi, python, typescript, fullstack]
+sources: [01 Projects/enterprise-ai-hub-nexus, 01 Projects/sandbox-notebookllamalm-nextjs, 01 Projects/Oliver-ai-bot_2.0, 01 Projects/ppt-tool]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# Next.js + FastAPI Fullstack
+
+Used for the most complex Oliver AI platforms. Next.js App Router handles SSR + auth; FastAPI handles AI workloads + data.
+
+## Key Takeaways
+- Next.js 14/15 App Router is the choice for production Oliver AI platforms
+- Paired with FastAPI backend (Python AI workloads, DB operations)
+- Azure AD PKCE auth lives in Next.js layer
+- Docker Compose orchestrates all services
+- Qdrant for vector search (RAG projects), Redis for caching/queuing
+
+## When to Use
+- Complex, multi-user AI platforms needing SSR, SEO, or M365 integration
+- When the frontend state is complex enough to warrant Next.js routing
+- Production deployments with multi-service architectures
+
+## Key Details
+- **Next.js versions:** 14 (Enterprise Nexus), 15 App Router (Sandbox NotebookLM)
+- **Package manager:** npm (standard), `uv` for Python backend (Sandbox)
+- **Auth:** Azure AD PKCE in Next.js; backend validates tokens
+- **API pattern:** `src/lib/api.ts` (axios wrapper) → FastAPI backend
+- **State:** Zustand (auth store in Sandbox NotebookLM)
+- **Real-time:** WebSocket chat routes in FastAPI (Sandbox NotebookLM)
+
+## Projects Using This Pattern
+- [[01 Projects/enterprise-ai-hub-nexus/Enterprise AI Hub Nexus|Enterprise AI Hub Nexus]] — Next.js 14 + FastAPI + PostgreSQL + Redis + Qdrant + Celery
+- [[01 Projects/sandbox-notebookllamalm-nextjs/Sandbox NotebookLM|Sandbox NotebookLM]] — Next.js 15 + FastAPI + Python 3.13 + uv + PostgreSQL + Redis + ElevenLabs
+- [[01 Projects/Oliver-ai-bot_2.0/Oliver AI Bot 2.0|Oliver AI Bot 2.0]] — Next.js + FastAPI (MVP, 85% backend complete)
+- [[01 Projects/ppt-tool/Oliver DeckForge|Oliver DeckForge]] — Next.js :3000 + FastAPI :8000 + nginx :80
+
+## Architecture Pattern
+```
+Browser → Azure AD PKCE (auth)
+    ↓
+Next.js (SSR + App Router)
+    src/lib/api.ts → FastAPI backend
+                       ├── AI service (Claude/Gemini/LlamaIndex)
+                       ├── PostgreSQL (data)
+                       ├── Redis (cache/queue)
+                       └── Celery workers (async jobs)
+```
+
+## Gotchas & Lessons
+- Next.js 15 App Router + React 19 is cutting edge — expect breaking changes in dependencies
+- `python-pptx` must be pinned to 0.6.23 for stable PPTX generation (Sandbox NotebookLM)
+- `max_tokens` defaults are too low for document generation — increase to 16000+ for long outputs
+- nginx reverse proxy needed when Next.js + FastAPI must share a single public port (DeckForge pattern)
+- Celery beat for scheduled tasks (SharePoint sync in Enterprise Nexus) needs careful token refresh logic
+
+## Related
+- [[wiki/tech-patterns/fastapi-python-docker|fastapi-python-docker]] — backend details
+- [[wiki/tech-patterns/azure-ad-msal-auth|azure-ad-msal-auth]] — auth layer
+- [[wiki/architecture/rag-architecture|rag-architecture]] — RAG pattern used by Nexus
+- [[wiki/tech-patterns/redis-celery-worker-queue|redis-celery-worker-queue]] — async processing
diff --git a/wiki/tech-patterns/nodejs-vanilla-proxy.md b/wiki/tech-patterns/nodejs-vanilla-proxy.md
new file mode 100644
index 0000000..3ef6dbc
--- /dev/null
+++ b/wiki/tech-patterns/nodejs-vanilla-proxy.md
@@ -0,0 +1,73 @@
+---
+title: "Node.js + Vanilla JS Lightweight Tools"
+aliases: [nodejs-proxy, vanilla-js, lightweight-tools]
+tags: [nodejs, vanilla-js, proxy, lightweight, no-build]
+sources: [01 Projects/3m-portal, 01 Projects/ferrero-ac-creator, 01 Projects/homepage]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# Node.js + Vanilla JS Lightweight Tools
+
+For simple client tools that don't need a frontend build step. Node.js serves static files and acts as a proxy for third-party APIs.
+
+## Key Takeaways
+- Use when: no React needed, no TypeScript, no build pipeline
+- `server.js` does two things: serve static HTML/JS/CSS files + proxy `/api/*` to third-party APIs
+- No Docker — just `node server.js` or `npm start`
+- No transpilation, no bundling — open file in browser or hit localhost directly
+- This pattern is for simple client-facing portals and utility tools
+
+## When to Use
+- Wrapping a third-party API that has CORS restrictions
+- Small client portals (translation tools, booking tools)
+- Rapid prototypes that don't need SSR or TypeScript
+
+## Key Details
+
+### server.js Structure
+```js
+const http = require('http')
+const fs = require('fs')
+const path = require('path')
+const https = require('https')
+
+const server = http.createServer((req, res) => {
+  if (req.url.startsWith('/api/')) {
+    // Proxy to third-party API
+    proxyRequest(req, res)
+  } else {
+    // Serve static files
+    serveStatic(req, res)
+  }
+})
+server.listen(3000)
+```
+
+### CORS Proxy Behavior (3M Portal)
+- Strips/rewrites `Location` headers on 301/302 → returns 401 (prevents auth redirect loops)
+- Injects CORS headers on all `/api` responses
+- Masks sensitive data (passwords) in logs
+
+## Projects Using This Pattern
+- [[01 Projects/3m-portal/3M OMG Portal|3M OMG Portal]] — Node.js CORS proxy for One2Edit API (:3000)
+- [[01 Projects/ferrero-ac-creator/Ferrero AC Booking Tool|Ferrero AC Booking]] — Node.js server for Box API + CSV download (:3456)
+- [[01 Projects/homepage/Homepage Dashboard|Homepage Dashboard]] — Node.js static server for YAML-configured dashboard
+
+## Comparison: When to Choose What
+| Need | Choose |
+|------|--------|
+| No CORS issues, pure static | Open `index.html` directly (Ferrero "Download CSV" mode) |
+| CORS proxy needed | `node server.js` |
+| Auth + complex logic | FastAPI (Python) |
+| SSR + complex routing | Next.js |
+
+## Gotchas & Lessons
+- `sessionStorage` (not `localStorage`) for auth tokens — clears on browser close (3M Portal)
+- Always mask passwords in proxy logs
+- 301/302 from API = auth failure in One2Edit — convert to 401 at proxy level
+- These tools have no test suite — test manually
+
+## Related
+- [[wiki/tech-patterns/one2edit-api|one2edit-api]] — uses this proxy pattern
+- [[wiki/tech-patterns/fastapi-python-docker|fastapi-python-docker]] — heavier alternative
diff --git a/wiki/tech-patterns/one2edit-api.md b/wiki/tech-patterns/one2edit-api.md
new file mode 100644
index 0000000..c7e3877
--- /dev/null
+++ b/wiki/tech-patterns/one2edit-api.md
@@ -0,0 +1,64 @@
+---
+title: "One2Edit API Integration"
+aliases: [one2edit, o2e]
+tags: [one2edit, api, translation, 3m, hm]
+sources: [01 Projects/3m-portal, 01 Projects/hm-o2e-tool]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# One2Edit API Integration
+
+One2Edit is an online editor/translation platform used by 3M and H&M for marketing document management. Oliver uses it at `oliver.one2edit.com`.
+
+## Key Takeaways
+- API endpoint: `https://oliver.one2edit.com/v3/Api.php`
+- Two auth modes: credential-based (service account) and session-based (externSessionId)
+- CORS proxy required — browsers can't call the API directly (same-origin policy)
+- The embedded editor uses the One2Edit JS SDK, not REST calls
+
+## When to Use
+Any client project built on the One2Edit platform (3M, H&M).
+
+## Key Details
+
+### Two Auth Modes
+| Mode | When | How |
+|------|------|-----|
+| Credential | Fetching job lists | Username → userId, then externSessionId |
+| Session | Embedded editor | `externSessionId` in SDK config |
+
+### 3M Portal Auth Flow
+```
+1. Login: username → POST /Api.php → userId
+2. Session: userId → POST /Api.php → externSessionId
+3. Dashboard: externSessionId → fetch jobs (STARTED/RUNNING)
+4. Editor: externSessionId → init One2Edit JS SDK
+```
+
+### CORS Proxy (3M Portal)
+```js
+// server.js
+// All /api/* requests → oliver.one2edit.com/v3/Api.php
+// Strips Location headers on 301/302 → returns 401
+// Injects CORS headers
+// Masks passwords in logs
+```
+
+### Service Account
+- 3M Portal uses `portal@oliver.agency` service account for job listing
+- Client users get their own `externSessionId` for the embedded editor
+
+## Projects Using This Pattern
+- [[01 Projects/3m-portal/3M OMG Portal|3M OMG Portal]] — Full portal: CORS proxy + Node.js backend + embedded SDK
+- [[01 Projects/hm-o2e-tool/HM O2E Tool|H&M O2E Tool]] — Static tool: image relinking + document export (no proxy needed — called directly or via `python -m http.server`)
+
+## Gotchas & Lessons
+- 301/302 redirects from One2Edit mean auth failure — the Node proxy converts them to 401 to prevent redirect loops in the browser
+- `sessionStorage` (not `localStorage`) — sessions clear on browser close, which is correct for this auth model
+- H&M O2E tool is static (no backend) — can run without a server for most operations
+- The JS SDK for the embedded editor is loaded from One2Edit's CDN — needs `externSessionId` at init time
+
+## Related
+- [[wiki/client-knowledge/hm|hm]] — H&M projects
+- [[wiki/tech-patterns/nodejs-vanilla-proxy|nodejs-vanilla-proxy]] — CORS proxy pattern
diff --git a/wiki/tech-patterns/python-ai-agents.md b/wiki/tech-patterns/python-ai-agents.md
new file mode 100644
index 0000000..a1284b7
--- /dev/null
+++ b/wiki/tech-patterns/python-ai-agents.md
@@ -0,0 +1,99 @@
+---
+title: "Python AI Integration Patterns"
+aliases: [ai-agents, claude-api, gemini-api, openai, llm-integration]
+tags: [ai, claude, gemini, openai, llamaindex, python, agents]
+sources: [01 Projects/gmal-scope-builder, 01 Projects/modcomms, 01 Projects/semblance, 01 Projects/enterprise-ai-hub-nexus, 01 Projects/olivas, 01 Projects/pimco-charts, 01 Projects/video-accessibility, 01 Projects/pdf-accessibility, 01 Projects/build-a-squad, 01 Projects/ac-helper]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# Python AI Integration Patterns
+
+How Oliver projects integrate LLMs — model selection, structured output, multi-model fallback.
+
+## Key Takeaways
+- **Claude** (Anthropic): `tool_use` for structured JSON output — used in GMAL, OliVAS, PIMCO, PDF Accessibility
+- **Gemini** (Google): dominant model for real-time/streaming analysis — Mod Comms, Video Accessibility, AC Helper, Build A Squad
+- **OpenAI**: used in Semblance (GPT-4.1, GPT-5.2) and Solventum
+- **LlamaIndex**: RAG orchestration in Sandbox NotebookLM (multi-model via `llm_factory.py`)
+- **DeepGaze**: specialized vision model for OliVAS (saliency/attention prediction)
+- Always implement a **fallback model**: Gemini Pro → Flash (Mod Comms pattern)
+
+## When to Use Which Model
+
+| Task | Model | Why |
+|------|-------|-----|
+| Structured JSON output | Claude (`tool_use`) | Reliable schema adherence |
+| Document analysis / image | Gemini 2.5 Pro | Multimodal, fast |
+| Real-time persona/focus groups | Gemini 3 Pro Preview / GPT-4.1 | Quality personas |
+| Metadata generation | OpenAI (Solventum) | Client preference |
+| RAG retrieval | LlamaIndex + any model | Framework abstraction |
+| Chart/SVG generation | Claude API | Precise code output |
+| Prompt optimization | Claude Sonnet 4.6 | Design analysis |
+| Video captions | Gemini 2.5 Pro | Native video understanding |
+
+## Structured Output — Claude `tool_use`
+```python
+# GMAL pattern: Claude tool_use for schema-bound output
+response = client.messages.create(
+    model="claude-opus-4-6",
+    tools=[{"name": "parse_brief", "input_schema": BriefSchema}],
+    tool_choice={"type": "tool", "name": "parse_brief"},
+    messages=[{"role": "user", "content": brief_text}]
+)
+result = response.content[0].input  # typed dict
+```
+
+## Multi-Model Fallback (Mod Comms Pattern)
+```python
+# Primary: Gemini Pro, fallback: Flash
+try:
+    result = await gemini_pro.generate(prompt)
+except Exception:
+    result = await gemini_flash.generate(prompt)
+```
+
+## LLM Factory Pattern (Sandbox NotebookLM)
+```python
+# llm_factory.py — abstract model selection
+def get_llm_by_type(llm_type: str) -> BaseLLM:
+    mapping = {"gemini-pro": GeminiPro(), "gpt4": OpenAI("gpt-4.1"), ...}
+    return mapping[llm_type]
+
+def get_structured_llm(schema: type) -> BaseLLM:
+    ...  # wraps output in Pydantic model
+```
+
+## AI Agents (Mod Comms Multi-Agent)
+```
+4 specialist agents (Legal, Brand, Tone, Channel)
+    → run in parallel
+    → Lead agent synthesizes verdict
+```
+See [[wiki/architecture/multi-agent-ai-systems|multi-agent-ai-systems]] for full pattern.
+
+## Projects Using This Pattern
+- [[01 Projects/gmal-scope-builder/GMAL Scope Builder|GMAL Scope Builder]] — Claude Opus 4.6, tool_use, structured ratecard output
+- [[01 Projects/modcomms/Mod Comms|Mod Comms]] — Gemini Pro + Flash fallback, 4 parallel agents
+- [[01 Projects/semblance/Semblance|Semblance]] — Gemini 3 Pro Preview, GPT-4.1, GPT-5.2, multi-persona
+- [[01 Projects/enterprise-ai-hub-nexus/Enterprise AI Hub Nexus|Enterprise Nexus]] — multi-model RAG via LlamaIndex + Qdrant
+- [[01 Projects/sandbox-notebookllamalm-nextjs/Sandbox NotebookLM|Sandbox NotebookLM]] — LlamaIndex, llm_factory, 7 studio generators
+- [[01 Projects/olivas/OliVAS|OliVAS]] — Claude Sonnet 4.6 design analysis + DeepGaze vision
+- [[01 Projects/pimco-charts/PIMCO Charts|PIMCO Charts]] — Claude API, SVG code generation
+- [[01 Projects/video-accessibility/Video Accessibility Platform|Video Accessibility]] — Gemini 2.5 Pro, VTT captions, audio description
+- [[01 Projects/pdf-accessibility/PDF Accessibility Checker|PDF Accessibility]] — Claude + Google Cloud Vision
+- [[01 Projects/build-a-squad/Build A Squad|Build A Squad]] — Gemini (client-side, no backend)
+- [[01 Projects/ac-helper/AC Helper|AC Helper]] — Gemini, natural language commands
+
+## Gotchas & Lessons
+- Long AI calls (>30s) will timeout on GCP LB — use HTTP polling, not streaming (see [[wiki/architecture/gcp-deployment-lb-timeout|gcp-deployment-lb-timeout]])
+- Increase Vite proxy timeout to 5 min for AI-heavy endpoints
+- Claude `tool_use` is more reliable than prompt-based JSON for structured output
+- `max_tokens` defaults (4096) too low for document generation — set 8192–16000
+- Gemini models change frequently — update model IDs in `llm_factory.py` regularly (Sandbox 2026-03-31)
+- GCP Vision + Claude hybrid (PDF Accessibility) — GCV handles image extraction, Claude does semantic analysis
+
+## Related
+- [[wiki/architecture/multi-agent-ai-systems|multi-agent-ai-systems]]
+- [[wiki/architecture/rag-architecture|rag-architecture]]
+- [[wiki/architecture/gcp-deployment-lb-timeout|gcp-deployment-lb-timeout]]
diff --git a/wiki/tech-patterns/react-vite-typescript.md b/wiki/tech-patterns/react-vite-typescript.md
new file mode 100644
index 0000000..9924374
--- /dev/null
+++ b/wiki/tech-patterns/react-vite-typescript.md
@@ -0,0 +1,63 @@
+---
+title: "React + Vite + TypeScript Frontend"
+aliases: [react-vite, frontend-stack]
+tags: [react, vite, typescript, frontend, tailwind, shadcn]
+sources: [01 Projects/gmal-scope-builder, 01 Projects/modcomms, 01 Projects/semblance, 01 Projects/video-accessibility, 01 Projects/lusa-back-planner, 01 Projects/build-a-squad, 01 Projects/baic_dashboard]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# React + Vite + TypeScript Frontend
+
+The standard Oliver frontend stack. React (18 or 19) + Vite + TypeScript, commonly paired with Tailwind CSS and shadcn/ui.
+
+## Key Takeaways
+- React 18 or 19 + Vite is the default for all new SPAs
+- Tailwind CSS (CDN for simple tools, npm for complex apps) + shadcn/ui for component-heavy UIs
+- `axios` for API calls; `react-router` for routing
+- For PDF: `pdfjs-dist` (parse) + `jsPDF` + `html2canvas` (export)
+- `shadcn/ui` (Radix UI + Tailwind) used in premium UI tools (Semblance, LUSA Back Planner)
+- Recharts for data visualization (BAIC Dashboard)
+
+## When to Use
+Any new Oliver UI that needs interactivity beyond a simple HTML page.
+
+## Key Details
+- **Port conventions:** 3000 (default), 3010 (GMAL), 5137 (Semblance), 5173 (Vite default)
+- **State management:** Zustand for auth (Sandbox NotebookLM) or local state
+- **API proxy:** Vite proxy to backend — increase timeout for AI calls (5 min)
+- **Build:** `npm run build` → `dist/` folder served by nginx or Node
+- **MSAL v5:** use `clearCache()` for logout (not `logout()` — breaking change in v5)
+
+## Styling Variants
+| Tool | Styling |
+|------|---------|
+| Semblance | Tailwind + shadcn/ui |
+| LUSA Back Planner | Tailwind + shadcn/ui |
+| GMAL Scope Builder | React + Vite + Axios |
+| Mod Comms | React 18 + Vite |
+| Video Accessibility | React 18 + Vite |
+| Build A Squad | React 19 + Vite |
+| BAIC Dashboard | React + Vite + Recharts |
+
+## Projects Using This Pattern
+- [[01 Projects/gmal-scope-builder/GMAL Scope Builder|GMAL Scope Builder]] — React 18 + TypeScript + Vite + React Router + Axios
+- [[01 Projects/modcomms/Mod Comms|Mod Comms]] — React 18 + Vite + TypeScript
+- [[01 Projects/semblance/Semblance|Semblance]] — React + TypeScript + Vite + Tailwind + shadcn/ui
+- [[01 Projects/video-accessibility/Video Accessibility Platform|Video Accessibility]] — React 18 + Vite + TypeScript
+- [[01 Projects/lusa-back-planner/LUSA Back Planner|LUSA Back Planner]] — React 19 + TypeScript + Vite + Tailwind + shadcn/ui + pdfjs + jsPDF
+- [[01 Projects/build-a-squad/Build A Squad|Build A Squad]] — React 19 + TypeScript + Vite + Gemini
+- [[01 Projects/baic_dashboard/BAIC Dashboard|BAIC Dashboard]] — React + Vite + Recharts + PapaParse
+- [[01 Projects/sandbox-notebookllamalm-nextjs/Sandbox NotebookLM|Sandbox NotebookLM]] — React 19 (inside Next.js 15 App Router)
+
+## Gotchas & Lessons
+- Vite proxy timeout defaults are too low for AI requests — set `proxy: { timeout: 300000 }` in vite.config
+- MSAL v5 breaking change: `clearCache()` not `logout()` for sign-out (GMAL fix 2026-03-30)
+- Send **ID token** (not Graph access token) to backend for Azure AD validation (GMAL fix 2026-03-30)
+- `shadcn/ui` components are copied into `src/components/ui/` — they're not an npm import at runtime
+- PapaParse for CSV export is lightweight and works browser-side
+
+## Related
+- [[wiki/tech-patterns/fastapi-python-docker|fastapi-python-docker]] — backend counterpart
+- [[wiki/tech-patterns/nextjs-fastapi-fullstack|nextjs-fastapi-fullstack]] — Next.js alternative
+- [[wiki/tech-patterns/azure-ad-msal-auth|azure-ad-msal-auth]] — auth integration
diff --git a/wiki/tech-patterns/redis-celery-worker-queue.md b/wiki/tech-patterns/redis-celery-worker-queue.md
new file mode 100644
index 0000000..bb7db4d
--- /dev/null
+++ b/wiki/tech-patterns/redis-celery-worker-queue.md
@@ -0,0 +1,92 @@
+---
+title: "Redis + Celery Async Worker Queue"
+aliases: [celery, task-queue, worker, redis-queue]
+tags: [redis, celery, async, worker, queue, python]
+sources: [01 Projects/enterprise-ai-hub-nexus, 01 Projects/video-accessibility, 01 Projects/pdf-accessibility]
+created: 2026-04-15
+updated: 2026-04-15
+---
+
+# Redis + Celery Async Worker Queue
+
+Pattern for offloading long-running AI/processing tasks to background workers. Used in the heaviest Oliver processing pipelines.
+
+## Key Takeaways
+- Redis is both the message broker AND result backend for Celery
+- Use Celery when tasks take >5s (AI inference, video processing, PDF analysis)
+- `Celery beat` for scheduled recurring tasks (e.g., SharePoint sync)
+- PDF Accessibility uses Redis queue directly (`pdf:queue`) without Celery — simpler `worker.py` daemon
+- Always poll for task status from the frontend; never block on long tasks
+
+## When to Use
+- Video processing pipelines (multi-phase, minutes-long)
+- Scheduled sync jobs (Celery beat)
+- Any task that would timeout an HTTP request (>30s)
+- Parallel AI analysis tasks
+
+## Key Details
+
+### Standard Setup
+```yaml
+# docker-compose.yml
+services:
+  redis:
+    image: redis:7
+    ports: ["6379:6379"]
+  worker:
+    build: ./backend
+    command: celery -A app.celery worker --loglevel=info
+    depends_on: [redis]
+  beat:
+    build: ./backend
+    command: celery -A app.celery beat --loglevel=info
+    depends_on: [redis]
+```
+
+### Task Definition
+```python
+@celery.task
+def process_video(video_id: str):
+    # Long-running pipeline
+    phase_1_ingest(video_id)
+    phase_2_caption(video_id)   # Gemini 2.5 Pro
+    phase_3_translate(video_id)
+    phase_4_tts(video_id)
+```
+
+### Polling Pattern (Frontend)
+```js
+// Poll until complete
+const poll = async (jobId) => {
+  const { status } = await api.get(`/jobs/${jobId}/status`)
+  if (status === 'pending') setTimeout(() => poll(jobId), 2000)
+}
+```
+
+## Projects Using This Pattern
+- [[01 Projects/enterprise-ai-hub-nexus/Enterprise AI Hub Nexus|Enterprise Nexus]] — Celery beat for SharePoint sync + scheduled tasks; Redis 7 + PostgreSQL
+- [[01 Projects/video-accessibility/Video Accessibility Platform|Video Accessibility]] — Celery workers for 7-phase video pipeline; Redis + MongoDB Atlas + GCS
+- [[01 Projects/pdf-accessibility/PDF Accessibility Checker|PDF Accessibility]] — Custom `worker.py` daemon reading `pdf:queue` from Redis; PostgreSQL for job tracking
+
+## Pipeline Phases (Video Accessibility)
+```
+1. Upload → Ingestion worker
+2. Gemini 2.5 Pro → VTT captions
+3. Audio Description generation
+4. QC review (approve/reject/edit VTT)
+5. Translation → 50+ languages
+6. TTS synthesis (GCP TTS + ElevenLabs)
+7. Final delivery
+```
+
+## Gotchas & Lessons
+- Celery beat needs its own container — it manages schedules independently from workers
+- Proactive token refresh required for long Celery jobs that need M365 access (Enterprise Nexus)
+- `worker.py` simpler alternative to Celery for single-queue use (PDF Accessibility pattern)
+- Always store job status in DB (not just Redis) so it survives Redis restart
+- `video_accessibility_development_plan.txt` is the authoritative spec — always read before touching that pipeline
+
+## Related
+- [[wiki/tech-patterns/fastapi-python-docker|fastapi-python-docker]] — the API layer above
+- [[wiki/tech-patterns/python-ai-agents|python-ai-agents]] — what the workers execute
+- [[wiki/architecture/gcp-deployment-lb-timeout|gcp-deployment-lb-timeout]] — why polling beats streaming