Chnaged requirement text

app/routes.py

@@ -4,6 +4,8 @@ import os
 import easyocr
 import pytesseract  # Ensure this is imported
 from PIL import Image
+
+from app.config import Config
 from app.models import audio_model, sentiment_pipeline, emotion_pipeline
 from app.services import extract_tasks
 from app.utils import generate_tags, error_response
@@ -13,11 +15,14 @@ bp = Blueprint('main', __name__)
 
 model_dir = os.getenv('EASYOCR_MODEL_STORAGE', None)
 
-# Initialize the EasyOCR reader for English only (disable GPU if not available)
-reader = easyocr.Reader(['en'], gpu=False
-,model_storage_directory=model_dir,
-user_network_directory=model_dir
-                        )
+# ── OCR via HF Inference API ─────────────────────────────────────────────────
+# We're using Microsoft's TrOCR for printed text:
+HF_API_TOKEN = Config.FIREWORKS_API_KEY
+ocr_api = InferenceApi(
+    repo_id="microsoft/trocr-base-printed",
+    token=HF_API_TOKEN,
+    inference_type="text-generation"  # TroCR is a seq2seq model
+)
 
 EMOTION_SCORE_THRESHOLD = 0.15  # Adjust based on your testing
 MIN_SENTIMENT_CONFIDENCE = 0.4  # Below this becomes "neutral"
@@ -64,33 +69,33 @@ def analyze_image():
         return error_response("No image file provided", 400)
 
     file = request.files['file']
-    filename = secure_filename(file.filename)
-    file_path = os.path.join("/tmp", filename)
-    file.save(file_path)
+    image_bytes = file.read()
 
     try:
-        # Use EasyOCR in detail mode to get confidence scores
-        results = reader.readtext(file_path, detail=1)
-
-        # Filter out entries with low confidence (e.g., below 0.5)
-        filtered_texts = [text for bbox, text, conf in results if conf > 0.5]
-        extracted_text = "\n".join(filtered_texts)
-
-        print("Filtered Extracted text:", extracted_text)
-
-        if not extracted_text.strip():
-            return error_response("No meaningful text found in image", 400)
-
-        # Analyze the extracted text to get sentiment, emotion, etc.
-        analysis_response = analyze_text_internal(extracted_text)
-        tags = generate_tags(extracted_text)
-
+        # send raw bytes to HF inference
+        result = ocr_api(image_bytes)
+        # TroCR returns a single string of text
+        extracted = ""
+        if isinstance(result, str):
+            extracted = result
+        elif isinstance(result, dict) and "generated_text" in result:
+            extracted = result["generated_text"]
+        else:
+            # fallback to printing whatever we got
+            extracted = str(result)
+
+        extracted = extracted.strip()
+        if not extracted:
+            return error_response("No text extracted from image", 400)
+
+        analysis = analyze_text_internal(extracted)
+        tags     = generate_tags(extracted)
         return jsonify({
-            "extracted_text": extracted_text.strip(),
-            "sentiment": analysis_response.get("sentiment"),
-            "emotion": analysis_response.get("emotion"),
-            "confidence": analysis_response.get("confidence"),
-            "tags": tags
+            "extracted_text": extracted,
+            "sentiment":     analysis["sentiment"],
+            "emotion":       analysis["emotion"],
+            "confidence":    analysis["confidence"],
+            "tags":          tags
         })
     except Exception as e:
         return error_response(str(e), 500)