[Edit] Key Rotate

llm.py

@@ -14,6 +14,8 @@ api_keys = [k.strip() for k in api_keys.split(",") if k.strip()]
 print(f"Loaded {len(api_keys)} Gemini API key(s)")
 
 def query_gemini(questions, contexts, max_retries=3):
+    import itertools
+
     context = "\n\n".join(contexts)
     questions_text = "\n".join([f"{i+1}. {q}" for i, q in enumerate(questions)])
 
@@ -77,43 +79,36 @@ Your task: For each question, provide a complete, professional, and clearly writ
 """
 
     last_exception = None
-
-    for key_idx, key in enumerate(api_keys):
-        for attempt in range(max_retries):
-            try:
-                genai.configure(api_key=key)
-                model = genai.GenerativeModel("gemini-2.5-flash-lite")
-                response = model.generate_content(prompt)
-                response_text = getattr(response, "text", "").strip()
-
-                if not response_text:
-                    raise ValueError("Empty response received from Gemini API.")
-
-                # Clean JSON from wrapped code blocks
-                if response_text.startswith("```json"):
-                    response_text = response_text.replace("```json", "").replace("```", "").strip()
-                elif response_text.startswith("```"):
-                    response_text = response_text.replace("```", "").strip()
-
-                # Optional debug log:
-                # print(f"[Gemini Response]: {response_text[:300]}")
-
-                parsed = json.loads(response_text)
-                if "answers" in parsed and isinstance(parsed["answers"], list):
-                    return parsed
-                else:
-                    raise ValueError("Invalid response format received from Gemini.")
-
-            except Exception as e:
-                msg = str(e).lower()
-                if "429" in msg or "quota" in msg or "rate limit" in msg or "exceeded" in msg:
-                    print(f"Gemini key {key[:8]}... quota/rate limited. Trying next key ({key_idx+1}/{len(api_keys)}).")
-                    last_exception = e
-                    break  # Try next key
-                else:
-                    print(f"Retry {attempt+1}/{max_retries} failed using Gemini key {key[:8]}...: {e}")
-                    last_exception = e
-                    continue  # Retry same key
-
-    print(f"All Gemini API keys failed. Last error: {last_exception}")
+    total_attempts = len(api_keys) * max_retries
+    key_cycle = itertools.cycle(api_keys)
+
+    for attempt in range(total_attempts):
+        key = next(key_cycle)
+        try:
+            genai.configure(api_key=key)
+            model = genai.GenerativeModel("gemini-2.5-flash-lite")
+            response = model.generate_content(prompt)
+            response_text = getattr(response, "text", "").strip()
+
+            if not response_text:
+                raise ValueError("Empty response received from Gemini API.")
+
+            if response_text.startswith("```json"):
+                response_text = response_text.replace("```json", "").replace("```", "").strip()
+            elif response_text.startswith("```"):
+                response_text = response_text.replace("```", "").strip()
+
+            parsed = json.loads(response_text)
+            if "answers" in parsed and isinstance(parsed["answers"], list):
+                return parsed
+            else:
+                raise ValueError("Invalid response format received from Gemini.")
+
+        except Exception as e:
+            last_exception = e
+            msg = str(e).lower()
+            print(f"[Retry {attempt+1}/{total_attempts}] Gemini key {key[:8]}... failed: {e}")
+            continue
+
+    print(f"All Gemini API attempts failed. Last error: {last_exception}")
     return {"answers": [f"Error generating response: {str(last_exception)}"] * len(questions)}