Upload app.py

app.py

@@ -303,8 +303,10 @@ def generate_prompt(query, context_chunks):
     """Generates a prompt for the LLM."""
     context_str = "\n\n".join(context_chunks)
     liaison_directory_url = "https://libguides.gc.cuny.edu/directory/subject"
-    prompt = f"""Based on the following context from the library guides, answer the user's question.
-If the context doesn't contain the answer, state that you couldn't find the information in the guides.
+    # Updated system prompt for clarity
+    prompt = f"""You are an AI assistant for the CUNY Graduate Center Library (also known as the Mina Rees Library).
+Based *only* on the following context extracted from the GC Library's LibGuides, answer the user's question about GC Library resources, services, or policies.
+Do not use any prior knowledge. If the context doesn't contain the answer, state that the information wasn't found in the provided LibGuides context.
 If your answer identifies a specific librarian or subject liaison, please also include this link to the main subject liaison directory: {liaison_directory_url}
 
 Context:
@@ -318,7 +320,7 @@ Answer:"""
     return prompt
 
 # --- Streamlit App UI ---
-st.title("📚 Experimental LibGuides RAG Demo") # Revised title
+st.title("📚 Ask the CUNY Graduate Center Library (RAG Demo)") # Updated title
 
 # User input (only proceed if collection loaded)
 if collection:
@@ -379,35 +381,38 @@ if collection and st.button("Ask"):
         st.warning("Please enter a question.")
     else:
         st.markdown("---")
-        with st.spinner("Routing query..."):
-            # --- LLM Routing Step ---
-            logging.info(f"Routing query: {query}")
-            routing_prompt = ROUTING_PROMPT_TEMPLATE.format(user_query=query)
-            try:
-                route_decision = query_hf_inference(routing_prompt).strip().upper()
-                logging.info(f"LLM (HF API) route decision: {route_decision}")
-                if route_decision.startswith("ERROR:"):
-                     st.error(f"Routing failed: {route_decision}")
-                     st.stop()
-            except Exception as e:
-                logging.error(f"LLM (HF API) routing failed: {e}. Defaulting to RAG.")
-                route_decision = "RAG"
-
-            # --- Handle specific routes ---
-            if route_decision == "HOURS":
-                st.info("You can find the current library hours here: [https://gc-cuny.libcal.com/hours](https://gc-cuny.libcal.com/hours)")
-                st.stop()
-            # ... (other routes) ...
-            elif route_decision == "EVENTS_CALENDAR":
-                events_url = "https://gc-cuny.libcal.com/calendar?cid=15537&t=d&d=0000-00-00&cal=15537&inc=0"
-                st.info(f"You can find information about upcoming library events and workshops on the calendar here: [{events_url}]({events_url})")
-                st.stop()
-            # --- End LLM Routing Step ---
+        # --- LLM Routing Step (Moved Before Spinner) ---
+        logging.info(f"Routing query: {query}")
+        routing_prompt = ROUTING_PROMPT_TEMPLATE.format(user_query=query)
+        try:
+            route_decision = query_hf_inference(routing_prompt).strip().upper()
+            logging.info(f"LLM (HF API) route decision: {route_decision}")
+            if route_decision.startswith("ERROR:"):
+                 st.error(f"Routing failed: {route_decision}")
+                 st.stop()
+        except Exception as e:
+            logging.error(f"LLM (HF API) routing failed: {e}. Defaulting to RAG.")
+            route_decision = "RAG" # Default to RAG on routing failure
 
-        spinner_text = "Thinking... (RAG)" if route_decision != "RESEARCH_QUERY" else "Thinking... (Research Query)"
-        with st.spinner(spinner_text):
-            # 1. Generate Query Variations (using HF API)
-            logging.info(f"Proceeding with retrieval for query (Route: {route_decision}): {query}")
+        # --- Handle specific routes immediately ---
+        if route_decision == "HOURS":
+            st.info("You can find the current library hours here: [https://gc-cuny.libcal.com/hours](https://gc-cuny.libcal.com/hours)")
+            st.stop()
+        elif route_decision == "EVENTS_CALENDAR":
+            events_url = "https://gc-cuny.libcal.com/calendar?cid=15537&t=d&d=0000-00-00&cal=15537&inc=0"
+            st.info(f"You can find information about upcoming library events and workshops on the calendar here: [{events_url}]({events_url})")
+            st.stop()
+        # Add other direct routes here if needed (e.g., CATALOG_SEARCH, ILL_REQUEST)
+        # elif route_decision == "CATALOG_SEARCH":
+        #     st.info("Search the library catalog here: [LINK_TO_CATALOG]")
+        #     st.stop()
+
+        # --- Proceed with RAG/Research Query if not handled above ---
+        if route_decision in ["RAG", "RESEARCH_QUERY"]: # Only proceed if it's a general or research query
+            spinner_text = "Thinking... (RAG)" if route_decision != "RESEARCH_QUERY" else "Thinking... (Research Query)"
+            with st.spinner(spinner_text):
+                # 1. Generate Query Variations (using HF API)
+                logging.info(f"Proceeding with retrieval for query (Route: {route_decision}): {query}")
             query_variations = generate_query_variations(query, query_hf_inference, HF_GENERATION_MODEL)
             all_queries = [query] + query_variations
             logging.info(f"--- DIAGNOSTIC: All queries for search: {all_queries}")
@@ -531,16 +536,16 @@ if collection and st.button("Ask"):
 # Add instructions or footer
 st.sidebar.header("About This Demo")
 st.sidebar.info(
-    "This is an experimental RAG demo using CUNY GC LibGuides content.\n\n"
-    "1. It loads pre-computed embeddings from a Hugging Face Dataset.\n"
-    "2. It embeds your query locally.\n"
-    "3. It uses the Hugging Face Inference API for LLM tasks (routing, generation).\n"
-    "4. Requires a `HUGGING_FACE_HUB_TOKEN` set as a Space secret (`HF_TOKEN`) or in `.env`."
+    "This is an experimental RAG demo for the CUNY Graduate Center Library (Mina Rees Library).\n\n"
+    "1. Loads pre-computed embeddings from a Hugging Face Dataset.\n"
+    "2. Embeds user queries locally.\n"
+    "3. Uses the Hugging Face Inference API for LLM generation.\n"
+    "4. Requires a `HUGGING_FACE_HUB_TOKEN` (set as Space secret `HF_TOKEN` or in `.env`)."
 )
 st.sidebar.header("Configuration Used")
-st.sidebar.markdown(f"**Embedding Source:** HF Dataset (`{HF_DATASET_ID}`)")
-st.sidebar.markdown(f"**Query Embedding Model:** Local (`{LOCAL_EMBEDDING_MODEL}`)")
-st.sidebar.markdown(f"**Generation LLM (HF API):** `{HF_GENERATION_MODEL}`")
+st.sidebar.markdown(f"**Data Source:** HF Dataset (`{HF_DATASET_ID}`)")
+st.sidebar.markdown(f"**Query Embedding:** Local (`{LOCAL_EMBEDDING_MODEL}`)")
+st.sidebar.markdown(f"**Generation LLM:** HF API (`{HF_GENERATION_MODEL}`)")
 st.sidebar.markdown(f"**Vector Store:** ChromaDB (In-Memory)")
 st.sidebar.markdown(f"**Retrieval Mode:** Vector Search Only")
 st.sidebar.markdown(f"**Final Unique Chunks:** `{TOP_K}` (from initial `{INITIAL_N_RESULTS}` vector search)")