Update app.py

app.py

@@ -18,7 +18,6 @@ from fastapi import FastAPI, File, UploadFile, Form, HTTPException, Security
 from fastapi.responses import JSONResponse
 from fastapi.security.api_key import APIKeyHeader
 
-import httpx
 from sarvamai import SarvamAI
 
 app = FastAPI(
@@ -319,7 +318,7 @@ def extract_text_with_sdk(file_bytes: bytes, filename: str) -> str:
                 pass
 
 # ---------------------------------------------------------------------------
-# Step 2: Extract structured JSON via Sarvam Chat (sarvam-m)
+# Step 2: Extract structured JSON via SarvamAI SDK chat completions
 # ---------------------------------------------------------------------------
 
 def strip_think_tags(text: str) -> str:
@@ -336,6 +335,9 @@ def strip_think_tags(text: str) -> str:
 
 
 async def extract_json_via_chat(doc_type: str, raw_text: str, parameters: list[str]) -> dict:
+    if not SARVAM_API_KEY:
+        raise HTTPException(status_code=500, detail="SARVAM_API_KEY is not configured on the server")
+
     description = SUPPORTED_TYPES[doc_type]
     param_list  = "\n".join(f"  - {p}" for p in parameters)
 
@@ -350,25 +352,25 @@ Rules:
 2. Use exactly the field names listed above as JSON keys.
 3. If a field is not found or not legible, set its value to null.
 4. Do not invent or infer values not explicitly present in the document.
-5. Resond everything in english language only. Also convert date into proper DD/MM/YYYY Foramt.
+5. Respond everything in english language only. Also convert date into proper DD/MM/YYYY format.
 Respond with the JSON object only."""
 
-    async with httpx.AsyncClient(timeout=60) as http:
-        resp = await http.post(
-            "https://api.sarvam.ai/v1/chat/completions",
-            headers={"api-subscription-key": SARVAM_API_KEY, "Content-Type": "application/json"},
-            json={
-                "model": "sarvam-30b",
-                "messages": [{"role": "user", "content": prompt}],
-                "max_tokens": 2048,
-                "temperature": 0
-            }
+    # Run the synchronous SDK call in a thread so we don't block the event loop
+    def _call_sdk() -> str:
+        client = SarvamAI(api_subscription_key=SARVAM_API_KEY)
+        response = client.chat.completions(
+            model="sarvam-30b",
+            messages=[{"role": "user", "content": prompt}],
+            temperature=0,
+            top_p=1,
+            max_tokens=2048,
         )
+        return response.choices[0].message.content.strip()
 
-    if resp.status_code != 200:
-        raise HTTPException(status_code=resp.status_code, detail=f"Sarvam chat error: {resp.text}")
-
-    raw = resp.json()["choices"][0]["message"]["content"].strip()
+    try:
+        raw = await asyncio.to_thread(_call_sdk)
+    except Exception as exc:
+        raise HTTPException(status_code=500, detail=f"Sarvam chat error: {exc}") from exc
 
     # Remove <think> blocks — handles both complete and truncated closing tags
     raw = strip_think_tags(raw)