feat(ml): configure Cloudinary upload for CDSCO scraper proof archiving

apps/api/src/routes/alerts.ts

@@ -18,6 +18,7 @@ const AlertSchema = z
         state: z.string().optional(),
         district: z.string().optional(),
         reported_at: z.string().optional(),
+        proof_image_url: z.string().optional().nullable(),
     })
     .passthrough();
 

apps/ml/requirements.txt

@@ -45,6 +45,7 @@ locust>=2.31.8
 # Agentic Orchestration and PDF rendering
 langgraph>=0.1.4
 pymupdf>=1.24.0
+cloudinary>=1.36.0
 
 # ScispaCy for Medical NER
 scispacy>=0.5.4

apps/ml/routers/asr.py

@@ -657,11 +657,7 @@ def _trim_audio_buffer(self) -> None:
         self.audio_buffer = self.audio_buffer[samples_to_trim:]
         self.buffer_start_seconds += samples_to_trim / STREAM_SAMPLE_RATE
 
-<<<<<<< HEAD
     def _build_response(self, transcript: str, *, run_ner: bool = False) -> dict:
-        base: dict = {
-=======
-    def _build_response(self, transcript: str) -> dict[str, str | float | bool | None]:
         medicine_db = get_medicine_database_list()
         fuzzy_match = get_phonetic_fuzzy_match(transcript, medicine_db)
 
@@ -674,8 +670,7 @@ def _build_response(self, transcript: str) -> dict[str, str | float | bool | Non
             suggestion_applied = True
             message = f"Showing results for {corrected_name} — did you mean this?"
 
-        return {
->>>>>>> pr-1840
+        base: dict = {
             "transcript": transcript,
             "corrected_name": corrected_name,
             "suggestion_applied": suggestion_applied,

apps/ml/services/alert_extractor.py

@@ -21,6 +21,7 @@ class AlertInfo(BaseModel):
     state: Optional[str] = Field(description="The state mentioned")
     district: Optional[str] = Field(description="The district mentioned")
     reported_at: Optional[str] = Field(description="The date of the alert in YYYY-MM-DD if possible")
+    proof_image_url: Optional[str] = Field(default=None, description="The Cloudinary URL of the original document snapshot proof")
 
 class AlertList(BaseModel):
     alerts: List[AlertInfo] = Field(description="List of alerts extracted from the text")
@@ -124,6 +125,7 @@ def extract_alerts_from_pdf_images(pdf_bytes: bytes) -> List[Dict[str, Any]]:
     """
     Extracts structured alert information from scanned/image-based PDFs
     by rendering pages to PNG images and analyzing them with Gemini-1.5-Flash.
+    Also uploads the page screenshots to Cloudinary to serve as audit proof.
     """
     if not LANGCHAIN_AVAILABLE:
         logging.error("LangChain dependencies missing.")
@@ -135,6 +137,29 @@ def extract_alerts_from_pdf_images(pdf_bytes: bytes) -> List[Dict[str, Any]]:
             logging.warning("GOOGLE_API_KEY not set. Cannot extract alerts using Gemini.")
             return []
 
+        # Configure Cloudinary if credentials are available
+        cloudinary_configured = False
+        cloud_name = os.getenv("CLOUDINARY_CLOUD_NAME")
+        api_key_cloud = os.getenv("CLOUDINARY_API_KEY")
+        api_secret_cloud = os.getenv("CLOUDINARY_API_SECRET")
+
+        if cloud_name and api_key_cloud and api_secret_cloud:
+            try:
+                import cloudinary
+                import cloudinary.uploader
+                cloudinary.config(
+                    cloud_name=cloud_name,
+                    api_key=api_key_cloud,
+                    api_secret=api_secret_cloud,
+                    secure=True
+                )
+                cloudinary_configured = True
+                logging.info("Cloudinary client configured successfully for CDSCO scraper proof archiving.")
+            except Exception as config_exc:
+                logging.warning(f"Failed to configure Cloudinary SDK: {config_exc}")
+        else:
+            logging.warning("Cloudinary environment variables missing, skipping proof uploads.")
+
         # Use fitz (PyMuPDF) to render PDF pages
         import fitz
         import base64
@@ -159,6 +184,28 @@ def extract_alerts_from_pdf_images(pdf_bytes: bytes) -> List[Dict[str, Any]]:
             png_bytes = pix.tobytes("png")
             b64_image = base64.b64encode(png_bytes).decode("utf-8")
 
+            # Upload the rendered page snapshot to Cloudinary if configured
+            proof_image_url = None
+            if cloudinary_configured:
+                try:
+                    import cloudinary.uploader
+                    import time
+                    timestamp = int(time.time())
+                    public_id = f"cdsco_alert_page_{page_num + 1}_{timestamp}"
+
+                    logging.info(f"Uploading page {page_num + 1} screenshot to Cloudinary as {public_id}...")
+                    upload_res = cloudinary.uploader.upload(
+                        png_bytes,
+                        folder="sahidawa/cdsco_proofs",
+                        public_id=public_id,
+                        overwrite=True,
+                        resource_type="image"
+                    )
+                    proof_image_url = upload_res.get("secure_url")
+                    logging.info(f"Cloudinary upload success: {proof_image_url}")
+                except Exception as upload_exc:
+                    logging.error(f"Cloudinary upload failed for page {page_num + 1}: {upload_exc}")
+
             # Prepare the multimodal prompt
             prompt_text = (
                 "You are an expert at extracting structured information from pharmaceutical recall and alert notices.\n"
@@ -205,6 +252,8 @@ def extract_alerts_from_pdf_images(pdf_bytes: bytes) -> List[Dict[str, Any]]:
                         )
                         if key not in seen_keys:
                             seen_keys.add(key)
+                            # Ingest the Cloudinary proof URL
+                            alert_dict["proof_image_url"] = proof_image_url
                             all_alerts.append(alert_dict)
             except Exception as page_exc:
                 logging.error(f"Error extracting alerts from page {page_num + 1} with Gemini: {page_exc}")

apps/ml/tests/test_alert_extractor.py

@@ -4,7 +4,8 @@
 
 @patch("services.alert_extractor.ChatGoogleGenerativeAI")
 @patch("fitz.open")
-def test_extract_alerts_from_pdf_images_success(mock_fitz_open, mock_chat_llm):
+@patch("cloudinary.uploader.upload")
+def test_extract_alerts_from_pdf_images_success(mock_cloudinary_upload, mock_fitz_open, mock_chat_llm):
     # Mock fitz document and pages
     mock_doc = MagicMock()
     mock_page = MagicMock()
@@ -37,13 +38,26 @@ def test_extract_alerts_from_pdf_images_success(mock_fitz_open, mock_chat_llm):
     mock_alert_list = AlertList(alerts=[mock_alert_info])
     mock_structured_llm.invoke.return_value = mock_alert_list
 
+    # Mock Cloudinary response
+    mock_cloudinary_upload.return_value = {"secure_url": "https://res.cloudinary.com/fake-url.png"}
+
     # Run extractor
     pdf_bytes = b"fake_pdf_content"
-    with patch.dict("os.environ", {"GOOGLE_API_KEY": "fake_key"}):
+    env_mock = {
+        "GOOGLE_API_KEY": "fake_key",
+        "CLOUDINARY_CLOUD_NAME": "fake_cloud",
+        "CLOUDINARY_API_KEY": "fake_api",
+        "CLOUDINARY_API_SECRET": "fake_secret"
+    }
+    with patch.dict("os.environ", env_mock):
         alerts = extract_alerts_from_pdf_images(pdf_bytes)
 
     assert len(alerts) == 1
     assert alerts[0]["reported_brand_name"] == "Mock Brand"
     assert alerts[0]["batch_number"] == "MB123"
     assert alerts[0]["manufacturer"] == "Mock Pharma"
     assert alerts[0]["alert_type"] == "NSQ"
+    assert alerts[0]["proof_image_url"] == "https://res.cloudinary.com/fake-url.png"
+
+    # Verify Cloudinary was called correctly
+    mock_cloudinary_upload.assert_called_once()

apps/ml/tests/test_asr_stream.py

@@ -115,6 +115,9 @@ def test_stream_returns_empty_final_when_stopped_before_audio():
         assert final == {
             "type": "final",
             "transcript": "",
+            "corrected_name": "",
+            "suggestion_applied": False,
+            "message": None,
             "language": None,
             "languageConfidence": None,
         }

supabase/migrations/20260612000000_add_drug_alerts_proof_url.sql

@@ -0,0 +1,2 @@
+-- Add proof_image_url column to drug_alerts table
+ALTER TABLE public.drug_alerts ADD COLUMN IF NOT EXISTS proof_image_url TEXT;