diff --git a/backend/app/services/embedding_service.py b/backend/app/services/embedding_service.py
index 8556945..ef2b1db 100644
--- a/backend/app/services/embedding_service.py
+++ b/backend/app/services/embedding_service.py
@@ -8,6 +8,7 @@ for memory and retry ergonomics with large glossaries).
 from __future__ import annotations
 
 import asyncio
+import re
 from collections.abc import Sequence
 
 from google import genai
@@ -20,8 +21,17 @@ logger = get_logger(__name__)
 
 _EMBED_MODEL = "gemini-embedding-001"
 _BATCH_SIZE = 100
-_MAX_RETRIES = 3
-_INITIAL_BACKOFF = 2.0
+_MAX_RETRIES = 5
+_INITIAL_BACKOFF = 8.0
+
+# Matches the 'retryDelay': '7s' field in Gemini 429 error bodies
+_RETRY_DELAY_RE = re.compile(r"'retryDelay':\s*'(\d+)s'")
+
+
+def _parse_retry_delay(exc: Exception) -> float | None:
+    """Extract the server-suggested retry delay from a Gemini 429 error."""
+    m = _RETRY_DELAY_RE.search(str(exc))
+    return float(m.group(1)) if m else None
 
 
 class EmbeddingService:
@@ -62,8 +72,12 @@ class EmbeddingService:
                 if attempt == _MAX_RETRIES:
                     logger.error(f"Embedding batch failed after {_MAX_RETRIES} attempts: {exc}")
                     raise
-                logger.warning(f"Embedding attempt {attempt} failed, retrying in {backoff}s: {exc}")
-                await asyncio.sleep(backoff)
+                # Honour the server-suggested retryDelay when present (e.g. 429 RESOURCE_EXHAUSTED).
+                # Fall back to our own exponential backoff otherwise.
+                server_delay = _parse_retry_delay(exc)
+                delay = max(server_delay + 1.0, backoff) if server_delay else backoff
+                logger.warning(f"Embedding attempt {attempt} failed, retrying in {delay}s: {exc}")
+                await asyncio.sleep(delay)
                 backoff *= 2
 
         raise RuntimeError("unreachable")  # makes type-checker happy
diff --git a/backend/app/tasks/embed_glossary.py b/backend/app/tasks/embed_glossary.py
index ba9941f..a2e49d0 100644
--- a/backend/app/tasks/embed_glossary.py
+++ b/backend/app/tasks/embed_glossary.py
@@ -20,7 +20,7 @@ from . import celery_app
 logger = get_logger(__name__)
 
 _BATCH_SIZE = 250
-_CONCURRENCY = 5
+_CONCURRENCY = 2
 
 
 @celery_app.task(name="embed_glossary_version", bind=True, max_retries=3)