feat: optimize auth concurrency and enable full account deletion

auth/routes.py

@@ -80,8 +80,18 @@ async def upload_avatar(
 
 @router.delete("/me")
 async def delete_account(user_id: str = Depends(get_current_user_id)):
+    # 1. Delete all DB data (materials, quizzes, profile, etc.)
     delete_user_data(user_id)
-    return {"status": "success", "message": "Account data deleted"}
+
+    # 2. Delete the Supabase Auth user so they can't sign in again
+    admin_client = get_supabase()
+    if admin_client:
+        try:
+            admin_client.auth.admin.delete_user(user_id)
+        except Exception as e:
+            raise HTTPException(500, f"Account data deleted but failed to remove auth user: {e}")
+
+    return {"status": "success", "message": "Account deleted successfully"}
 
 
 @router.get("/profile")

rag/constants.py

@@ -6,15 +6,15 @@ WARMUP_INTERVAL_S = 300
 
 # Web search configuration — Wiki + DDG for topics, DDG only for PDF/URL materials.
 WIKI_TOP_K_RESULTS = 1                   # Number of top Wikipedia articles retrieved
-WIKI_DOC_CONTENT_CHARS_MAX = 1200        # Max chars from Wikipedia result
+WIKI_DOC_CONTENT_CHARS_MAX = 1000         # Max chars from Wikipedia result
 
 # DuckDuckGO Search
-DUCKDUCKGO_NUM_RESULTS = 4               # Number of DDG snippet results returned per search
-DUCKDUCKGO_DOC_CONTENT_CHARS_MAX = 2500  # Max chars kept from combined DDG result block
+DUCKDUCKGO_NUM_RESULTS = 3               # Number of DDG snippet results returned per search
+DUCKDUCKGO_DOC_CONTENT_CHARS_MAX = 1500  # Max chars kept from combined DDG result block
 
 # RAG & Memory Configuration
-MEMORY_WINDOW_SIZE = 12                   # Number of previous conversation turns preserved in memory window
-TOP_K_CHUNKS = 4                         # Number of top relevant material chunks retrieved for context
+MEMORY_WINDOW_SIZE = 8                   # Number of previous conversation turns preserved in memory window
+TOP_K_CHUNKS = 3                         # Number of top relevant material chunks retrieved for context
 
 RAG_PROMPT_TEMPLATE_BASE = """\
 <role>

rag/rag.py

@@ -386,7 +386,7 @@ def rag_answer(
         return any(t.startswith(p) for p in _REFUSAL_PREFIXES)
 
     try:
-        primary_llm = get_gemma_31b_llm()
+        primary_llm = get_gemma_26b_llm()
         chain = prompt | primary_llm
 
         memory_vars = memory.load_memory_variables({"input": query})
@@ -405,9 +405,9 @@ def rag_answer(
             memory.save_context({"input": query}, {"output": answer})
         return answer, memory
     except Exception as e:
-        logger.warning(f"Gemma 4 31B API call failed or rate-limited: {e}. Falling back to gemma-4-26b-a4b-it immediately.")
+        logger.warning(f"Gemma 4 26B API call failed or rate-limited: {e}. Falling back to gemma-4-31b-it immediately.")
         try:
-            fallback_llm = get_gemma_26b_llm()
+            fallback_llm = get_gemma_31b_llm()
             chain = prompt | fallback_llm
             memory_vars = memory.load_memory_variables({"input": query})
             chat_history = memory_vars.get("chat_history", [])
@@ -424,7 +424,7 @@ def rag_answer(
                 memory.save_context({"input": query}, {"output": answer})
             return answer, memory
         except Exception as fallback_err:
-            logger.error(f"Fallback Gemma 4 26B LLM call also failed: {fallback_err}")
+            logger.error(f"Fallback Gemma 4 31B LLM call also failed: {fallback_err}")
             raise fallback_err
 
 def extract_chat_title(query: str, material_title: Optional[str] = None) -> str:
@@ -440,17 +440,17 @@ def extract_chat_title(query: str, material_title: Optional[str] = None) -> str:
     )
     
     try:
-        primary_llm = get_gemma_31b_llm()
+        primary_llm = get_gemma_26b_llm()
         chain = prompt | primary_llm
         response = chain.invoke({"query": query})
     except Exception as e:
-        logger.warning(f"Gemma 4 31B API call failed or rate-limited in extract_chat_title: {e}. Falling back to gemma-4-26b-a4b-it immediately.")
+        logger.warning(f"Gemma 4 26B API call failed or rate-limited in extract_chat_title: {e}. Falling back to gemma-4-31b-it immediately.")
         try:
-            fallback_llm = get_gemma_26b_llm()
+            fallback_llm = get_gemma_31b_llm()
             chain = prompt | fallback_llm
             response = chain.invoke({"query": query})
         except Exception as fallback_err:
-            logger.error(f"Fallback Gemma 4 26B LLM call also failed in extract_chat_title: {fallback_err}")
+            logger.error(f"Fallback Gemma 4 31B LLM call also failed in extract_chat_title: {fallback_err}")
             raise fallback_err
 
     title = _clean_llm_response(response.content).strip().strip('"').strip("'")