Add backend enhancements: AsyncOpenAI, image compression, refined prompts, and logging

requirements.txt

@@ -4,3 +4,4 @@ openai
 python-dotenv
 python-multipart
 pydantic
+Pillow

src/services/openai_service.py

@@ -1,19 +1,55 @@
 import os
 import base64
-from openai import OpenAI
+import io
+import logging
+from openai import AsyncOpenAI
 from dotenv import load_dotenv
+from PIL import Image
 from src.schemas import ExtractionResponse
 
 load_dotenv()
 
-client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
+# Setup logging
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
+
+client = AsyncOpenAI(api_key=os.getenv("OPENAI_API_KEY"))
+
+MAX_IMAGE_SIZE = (2000, 2000)
+IMAGE_QUALITY = 85
+
+def compress_image(image_content: bytes) -> bytes:
+    """Resizes and compresses the image to reduce upload size."""
+    img = Image.open(io.BytesIO(image_content))
+    
+    # Convert to RGB if necessary (to save as JPEG)
+    if img.mode in ("RGBA", "P"):
+        img = img.convert("RGB")
+    
+    # Resize if larger than max dimensions
+    img.thumbnail(MAX_IMAGE_SIZE, Image.Resampling.LANCZOS)
+    
+    # Save to bytes
+    output_buffer = io.BytesIO()
+    img.save(output_buffer, format="JPEG", quality=IMAGE_QUALITY, optimize=True)
+    return output_buffer.getvalue()
 
 async def extract_receipt_data(image_content: bytes, user_name: str, department: str) -> ExtractionResponse:
-    base64_image = base64.b64encode(image_content).decode("utf-8")
+    # 1. Compress Image
+    compressed_content = compress_image(image_content)
+    base64_image = base64.b64encode(compressed_content).decode("utf-8")
     
-    prompt = f"You are an HR data entry assistant. Extract the requested fields from the provided medical receipt image. The employee submitting this is {user_name} from {department}. If the date is missing, look for a 'Payment Date' as a fallback."
+    # 2. Refined Prompt
+    prompt = (
+        f"You are an HR data entry assistant. Extract the requested fields from the provided medical receipt image. "
+        f"The employee submitting this is {user_name} from {department}. "
+        f"If the date is missing, look for a 'Payment Date' as a fallback. "
+        f"Analyze the receipt for authenticity. If the total amount appears altered or if the provider name is missing, "
+        f"set `needs_manual_review` to `true` and provide a low `confidence_score`."
+    )
     
-    completion = client.beta.chat.completions.parse(
+    # 3. Async Extraction
+    completion = await client.beta.chat.completions.parse(
         model="gpt-4o",
         messages=[
             {
@@ -30,4 +66,12 @@ async def extract_receipt_data(image_content: bytes, user_name: str, department:
         response_format=ExtractionResponse,
     )
     
-    return completion.choices[0].message.parsed
+    result = completion.choices[0].message.parsed
+    
+    # 4. Logging for Demo
+    if result:
+        logger.info(f"Extraction complete for {user_name}. Confidence Score: {result.confidence_score}")
+        if result.needs_manual_review:
+            logger.warning(f"Manual review required for receipt submitted by {user_name}")
+            
+    return result