fix: remove failing Groq Vision API, use only free OCR.space API for scanned document OCR

backend/parsers/llm_summarizer.py

@@ -311,61 +311,10 @@ class LLMSummarizer:
                 return await self._summarize_with_text(minutes)
             raise RuntimeError("No image data and no text fallback.")
 
-        # Try free Groq Vision API (Llama 3.2 11B Vision) to read text from images.
-        # No JSON mode needed — just raw text extraction. Then text model summarizes.
-        import base64
-        print("[OCR] Using Groq Vision API to read text from images...")
-        vision_text = ""
-        try:
-            max_pages = min(len(image_bytes_list), 3)
-            content = [{"type": "text", "text": "Extract ALL readable text from these scanned city council meeting pages. Transcribe everything exactly as you see it. Preserve the layout."}]
-            for i in range(max_pages):
-                img_b64 = base64.b64encode(image_bytes_list[i]).decode("utf-8")
-                mime = "image/png" if image_bytes_list[i][:4] == b'\x89PNG' else "image/jpeg"
-                content.append({"type": "image_url", "image_url": {"url": f"data:{mime};base64,{img_b64}"}})
-            vr = self.deepseek_client.chat.completions.create(
-                model="llama-3.2-11b-vision-preview",
-                messages=[{"role": "user", "content": content}],
-                temperature=0.0, max_tokens=4096,
-            )
-            vision_text = vr.choices[0].message.content or ""
-            print(f"[VISION] Extracted {len(vision_text)} chars from {max_pages} pages")
-            print(f"[VISION] First 200 chars: {vision_text[:200]}")
-        except Exception as e:
-            print(f"[VISION] Failed: {e}")
-
-        if vision_text and len(vision_text.strip()) > 50:
-            minutes.raw_text = vision_text.strip()
-            user_content = (
-                f"Please summarize the following city council meeting minutes "
-                f"for {minutes.city}, {minutes.state} on "
-                f"{minutes.meeting_date.strftime('%B %d, %Y')}.\n\n"
-                f"Meeting Type: {minutes.meeting_type}\n"
-                f"Title: {minutes.title}\n\n"
-                f"The following text was extracted from scanned document images by an AI vision model:\n\n"
-                f"{vision_text.strip()[:self.MAX_OCR_TEXT_CHARS]}\n\n"
-                f"Return ONLY valid JSON matching the specified structure. "
-                f"Do not include any text outside the JSON object."
-            )
-            response = self.deepseek_client.chat.completions.create(
-                model=self.text_model,
-                messages=[{"role": "system", "content": MINUTES_SYSTEM_PROMPT}, {"role": "user", "content": user_content}],
-                temperature=0.0, max_tokens=4096, response_format={"type": "json_object"},
-            )
-            result = json.loads(response.choices[0].message.content)
-            return SummaryResponse(
-                minutes_id=minutes.id, meeting_date=minutes.meeting_date, meeting_type=minutes.meeting_type,
-                big_picture=result.get("big_picture", ""),
-                summary=result.get("summary", "No summary available."),
-                key_decisions=[d if isinstance(d, dict) else {"description": d} for d in result.get("key_decisions", [])],
-                budget_items=[d if isinstance(d, dict) else {"description": d} for d in result.get("budget_items", [])],
-                public_comment_opportunities=[d if isinstance(d, dict) else {"description": d} for d in result.get("public_comment_opportunities", [])],
-                items=[d if isinstance(d, dict) else {"description": d} for d in result.get("items", [])],
-                what_you_can_do=[d if isinstance(d, dict) else {"description": d} for d in result.get("what_you_can_do", [])],
-            )
-
-        # Fallback: OCR.space (synchronous API call)
-        print("[OCR] Vision failed, trying free OCR.space API")
+        # Use free OCR.space API to extract text from scanned images.
+        # Works on Vercel Lambda with no local binaries needed.
+        # Rate limit: 25k requests/month free tier, no API key required.
+        print("[OCR] Extracting text via free OCR.space API...")
         ocr_parts = []
         for i, img_bytes in enumerate(image_bytes_list[:5]):
             page_text = self._ocr_with_ocrspace(img_bytes)