Update app.py

app.py

@@ -29,31 +29,48 @@ llm = AutoModelForCausalLM.from_pretrained(llama_model_name, token=API_TOKEN)
 # Generate advice using RAG
 def generate_advice(extracted_data):
     try:
+        # Ensure extracted_data is valid
+        if not isinstance(extracted_data, list):
+            raise ValueError("Input data must be a list of dictionaries.")
+        if not all(isinstance(item, dict) for item in extracted_data):
+            raise ValueError("Each item in input data must be a dictionary.")
+
         recommendations = []
 
         for item in extracted_data:
+            # Validate required keys
+            if not all(k in item for k in ["Component", "Status"]):
+                raise ValueError("Each input item must have 'Component' and 'Status' keys.")
+            
             # Prepare the query string
             query = f"{item['Component']} {item['Status']}"
             print(f"Processing query: {query}")  # Debug print
 
-            # Generate query embedding and ensure it’s a 2D numpy array
+            # Generate query embedding and reshape
             query_embedding = embedding_model.encode([query])
             query_embedding = np.array(query_embedding, dtype="float32").reshape(1, -1)
 
+            # Debugging embedding dimensions
+            print(f"Query Embedding Shape: {query_embedding.shape}, FAISS Index Dim: {index.d}")
+
+            # Validate embedding dimensions
+            if query_embedding.shape[1] != index.d:
+                raise ValueError(
+                    f"Embedding dimension mismatch: Query ({query_embedding.shape[1]}), Index ({index.d})"
+                )
+
             # Search for the closest match in FAISS
             _, idx = index.search(query_embedding, 1)
-            print(f"FAISS Index: {idx}, Best Match Raw: {kb[idx[0][0]]}")  # Debug print
+            print(f"FAISS Index: {idx}, Best Match Raw: {kb[idx[0][0]]}")
 
-            # Retrieve the closest match and validate
+            # Retrieve the closest match
             best_match = kb[idx[0][0]]
-            if not isinstance(best_match, dict):
-                raise ValueError(f"Best match retrieved is not a dictionary: {best_match}")
 
             # Prepare the LLM prompt
             role = "Medical expert providing advice based on lab results."
             prompt = f"""
             Lab Test: {item['Component']}
-            Value: {item['Value']} {item['Units']}
+            Value: {item.get('Value', 'Unknown')} {item.get('Units', '')}
             Status: {item['Status']}
             
             Medical Guidelines: {best_match['Advice']}
@@ -61,6 +78,7 @@ def generate_advice(extracted_data):
             Provide additional insights or recommendations.
             """
 
+            # Generate advice using LLaMA model
             message_yours = [
                 {"role": "system", "content": role},
                 {"role": "user", "content": prompt},
@@ -73,7 +91,6 @@ def generate_advice(extracted_data):
                 return_tensors="pt",
             )
 
-            # Generate response using LLaMA
             output = llm.generate(
                 input_ids=input_text_with_your_role,
                 max_length=150,
@@ -81,7 +98,7 @@ def generate_advice(extracted_data):
             )
             advice = tokenizer.decode(output[0], skip_special_tokens=True).strip()
 
-            # Append result
+            # Append the result
             recommendations.append({
                 "Component": item["Component"],
                 "Advice": advice