feat: increase quiz generation capacity with dedicated high-token LLM.

auth/routes.py

@@ -94,7 +94,11 @@ async def get_profile(
                 "email": getattr(user_obj, "email", "") or "",
                 "avatar_url": "",
                 "daily_requests": real_usage.get("used", 0),
-                "last_request_date": __import__('datetime').date.today().isoformat(),
+                "last_request_date": (
+                    __import__('datetime').datetime.now(
+                        __import__('datetime').timezone(__import__('datetime').timedelta(hours=3))
+                    ).date().isoformat()
+                ),
                 "_is_fallback": True,
             })
 

quiz_generator/constants.py

@@ -1,7 +1,7 @@
 from langchain.prompts import PromptTemplate
 
 MIN_MCQ_COUNT = 1
-MAX_MCQ_COUNT = 20
+MAX_MCQ_COUNT = 40
 MIN_TF_COUNT = 1
 MAX_TF_COUNT = 20
 

quiz_generator/quiz.py

@@ -6,7 +6,7 @@ from typing import Optional
 from langchain.agents import create_tool_calling_agent, AgentExecutor
 from langchain_core.tools import create_retriever_tool
 
-from src.rag.rag import get_llm, web_search_tools, SupabaseRetriever
+from src.rag.rag import get_quiz_llm, web_search_tools, SupabaseRetriever
 from .constants import (
     QUIZ_PROMPT_TEMPLATE,
     WEB_QUIZ_PROMPT_TEMPLATE,
@@ -60,7 +60,7 @@ def _summary_quiz(difficulty, mcq_count, tf_count, context_text):
     logger.info(f"Summary Quiz started (diff={difficulty}, mcq={mcq_count}, tf={tf_count})")
     try:
         prompt = _quiz_prompt()
-        llm = get_llm()
+        llm = get_quiz_llm()
         safe_context = context_text
         
         chain = prompt | llm
@@ -87,7 +87,7 @@ def _contextual_quiz(difficulty, mcq_count, tf_count, context, material_id):
     logger.info(f"Contextual Quiz started (material_id={material_id}, diff={difficulty})")
     try:
         prompt = _quiz_prompt()
-        llm = get_llm()
+        llm = get_quiz_llm()
 
         retriever = SupabaseRetriever(material_id=material_id, k=RETRIEVER_K)
         retriever_tool = create_retriever_tool(
@@ -103,6 +103,8 @@ def _contextual_quiz(difficulty, mcq_count, tf_count, context, material_id):
             verbose=False,
             return_intermediate_steps=False,
             handle_parsing_errors=True,
+            max_iterations=80,
+            max_execution_time=300,
         )
 
         safe_context = context or ""
@@ -125,7 +127,7 @@ def _web_quiz(difficulty, mcq_count, tf_count, topic_title):
     logger.info(f"Web Quiz started (topic={topic_title}, diff={difficulty})")
     try:
         prompt = WEB_QUIZ_PROMPT_TEMPLATE
-        llm = get_llm()
+        llm = get_quiz_llm()
         tools = web_search_tools(
             wiki_k=WIKI_TOP_K_RESULTS,
             wiki_chars=WIKI_DOC_CONTENT_CHARS_MAX,
@@ -140,6 +142,8 @@ def _web_quiz(difficulty, mcq_count, tf_count, topic_title):
             verbose=False,
             return_intermediate_steps=False,
             handle_parsing_errors=True,
+            max_iterations=80,
+            max_execution_time=300,
         )
 
         safe_context = topic_title

quiz_generator/schemas.py

@@ -3,8 +3,8 @@ from typing import Optional
 
 class QuizRequest(BaseModel):
     difficulty: str = "Medium"
-    mcq_count: int = 4
-    tf_count: int = 3
+    mcq_count: int = 10
+    tf_count: int = 5
     source_type: str = "web"
     material_id: Optional[str] = None
     topic: Optional[str] = None

rag/rag.py

@@ -146,11 +146,28 @@ def get_llm():
         model=settings.model_name,
         api_key=settings.gemini_api_key,
         temperature=0.3,
-        max_output_tokens=2000,
+        max_output_tokens=2500,
         timeout=120,
     )
 
 
+def get_quiz_llm():
+    """Dedicated LLM instance for quiz generation with a high output token budget.
+    Large quizzes (40 MCQs + 20 T/F) can produce 8000-12000 tokens of JSON,
+    so we cannot reuse the chat LLM which is capped at 2000 tokens.
+    """
+    if not os.environ.get("GEMINI_API_KEY"):
+        raise ValueError("GEMINI_API_KEY not found. Please set it in config.env.")
+    logger.info(f"Initializing Quiz LLM with model: {settings.model_name}")
+    return ChatGoogleGenerativeAI(
+        model=settings.model_name,
+        api_key=settings.gemini_api_key,
+        temperature=0.3,
+        max_output_tokens=16000,
+        timeout=300,
+    )
+
+
 def get_groq_llm():
     if not settings.groq_api_key:
         raise ValueError("GROQ_API_KEY not found. Please set it in config.env.")

store.py

@@ -3,10 +3,15 @@ import time
 from httpx import RemoteProtocolError
 from typing import Optional
 import logging
-from datetime import datetime, timezone, date
+from datetime import datetime, timezone, date, timedelta
 from langchain.memory import ConversationBufferMemory, ConversationBufferWindowMemory
 from src.database import get_supabase
 
+def _get_today_date_str() -> str:
+    # Shift UTC time by 3 hours to match Egypt timezone (UTC+3), so daily limits reset at 12 AM Egypt time.
+    egypt_tz = timezone(timedelta(hours=3))
+    return datetime.now(egypt_tz).date().isoformat()
+
 _in_memory: dict = {
     "materials": {},
     "material_chunks": {},
@@ -384,7 +389,7 @@ def get_quizzes(material_id: Optional[str] = None, user_id: Optional[str] = None
 # ── Users (maps to Supabase `profiles` table) ─────────
 
 def _map_profile(profile: dict) -> dict:
-    today = date.today().isoformat()
+    today = _get_today_date_str()
     used = profile.get("daily_requests", 0) if profile.get("last_request_date") == today else 0
     return {
         "id": profile["id"],
@@ -646,32 +651,15 @@ def check_and_increment_daily_limit(user_id: str, email: Optional[str] = None, l
     Returns True if request is allowed, False if limit exceeded.
     Excludes Admin Emails from Limits.
 
-    Uses a single atomic SQL UPDATE via Supabase RPC to eliminate the
-    check-then-increment race condition (two concurrent reads both see
-    count=9, both pass, both increment → 11th request gets through).
+    Bypasses the database RPC (which runs in UTC and resets at 3 AM Egypt time)
+    to perform the date check in python using Egypt local timezone (UTC+3), 
+    ensuring limits reset exactly at 12 AM Egypt time.
     """
     # Guard: never apply limit to admin emails
     if email and email in ADMIN_EMAILS:
         return True
 
-    client = _db()
-    if client is not None:
-        try:
-            result = client.rpc(
-                "increment_daily_limit",
-                {"p_user_id": user_id, "p_limit": limit},
-            ).execute()
-            if not result.data:
-                return False
-            return True
-        except Exception as e:
-            logger.warning(f"Atomic rate-limit RPC failed, falling back to two-query path: {e}")
-            # Fall through to the two-query fallback below
-
-    # ── Offline / dev fallback (also used when RPC call above raises) ──────────
-    # This path has a theoretical race condition but is acceptable for single-
-    # worker dev environments where the RPC is not available.
-    today = date.today().isoformat()
+    today = _get_today_date_str()
     try:
         result = _robust_execute(
             _table_supabase("profiles")
@@ -706,7 +694,7 @@ def get_usage(user_id: str) -> dict:
     """
     Returns current usage for a user.
     """
-    today = date.today().isoformat()
+    today = _get_today_date_str()
     try:
         result = _robust_execute(
             _table_supabase("profiles")

summary_generator/constants.py

@@ -104,7 +104,6 @@ Analyze the content and produce a summary with exactly these five sections in th
    - **Historical Background** (e.g., [[[[>>> Historical Background <<<]]]]) — Brief, concise history: when it originated, key milestones, and major contributors. Keep this summarized.
    - **Core Concepts and Fundamentals** (e.g., [[[[>>> Core Concepts and Fundamentals <<<]]]]) — The essential principles, mechanisms, theories, or ideas that form the foundation of this topic.
    - **Types / Categories / Variants** (e.g., [[[[>>> Types and Classifications <<<]]]]) — If the topic has distinct types, classifications, branches, or variants, list and briefly explain each one.
-   - **Architecture / Structure / Components** (e.g., [[[[>>> Architecture and Components <<<]]]]) — If applicable, describe the internal structure, architecture, system design, or key components and how they relate.
    - **How It Works / Process / Mechanism** (e.g., [[[[>>> How It Works <<<]]]]) — Step-by-step explanation of how it functions, operates, or proceeds, if applicable.
    - **Applications and Use Cases** (e.g., [[[[>>> Applications and Use Cases <<<]]]]) — Real-world applications, practical uses, and examples of where this topic is applied.
    - **Advantages and Strengths** (e.g., [[[[>>> Advantages and Strengths <<<]]]]) — Key benefits, strengths, and reasons why this topic/approach is valuable.