Update app.py

app.py

@@ -39,29 +39,30 @@ def detect_query(query):
 # -----------------------------
 # Retrieve context (RAG)
 # -----------------------------
-def retrieve_context(query):
+def retrieve_context(query, top_k=2):
     animal, topic = detect_query(query)
 
-    filtered_indices = []
-    for i, meta in enumerate(metadata):
-        if animal and meta["animal"] != animal:
-            continue
-        if topic and meta["topic"] != topic:
-            continue
-        filtered_indices.append(i)
+    # Filter relevant chunks based on metadata
+    filtered_indices = [
+        i for i, meta in enumerate(metadata)
+        if (not animal or meta["animal"] == animal) and
+           (not topic or meta["topic"] == topic)
+    ]
 
+    # If no specific filter matches, consider all chunks
     if not filtered_indices:
         filtered_indices = list(range(len(chunks)))
 
+    # Embed query
     query_embedding = embed_model.encode([query])
     filtered_embeddings = np.array([index.reconstruct(i) for i in filtered_indices])
+
+    # Compute distances and get top-k closest chunks
     distances = np.linalg.norm(filtered_embeddings - query_embedding, axis=1)
-    top_indices = distances.argsort()[:2]
+    top_indices = distances.argsort()[:top_k]
 
-    context = ""
-    for idx in top_indices:
-        real_index = filtered_indices[idx]
-        context += chunks[real_index] + "\n"
+    # Combine top chunks into context
+    context = "\n".join(chunks[filtered_indices[idx]] for idx in top_indices)
 
     return context.strip()
 
@@ -72,15 +73,18 @@ def chat(user_input):
     context = retrieve_context(user_input)
     if not context:
         return "I don't know."
-    return context
+    
+    # Return context with clear formatting
+    return f"Answer from retrieved data:\n\n{context}"
 
 # -----------------------------
 # Gradio UI
 # -----------------------------
 gr.Interface(
     fn=chat,
-    inputs="text",
-    outputs="text",
+    inputs=gr.Textbox(lines=2, placeholder="Ask a question about livestock..."),
+    outputs=gr.Textbox(),
     title="Livestock Chatbot (RAG only)",
-    description="This chatbot answers livestock questions using only the retrieved data. No AI model is used."
+    description="This chatbot answers livestock questions using only retrieved data. No AI model is used.",
+    allow_flagging="never"
 ).launch()