update welcom message + fix clarification/search prompts to account for sql results

app_web.py

@@ -357,7 +357,12 @@ def process_contact_request(original_query:str, qcat_result:Dict, uploaded_files
             
     with st.chat_message("assistant"):
         st.markdown(response)       
-             
+        feedback = st.feedback(options="faces", key=f"fb_contact_{st.session_state.resolution_rounds-1}")
+
+    st.session_state.messages.append({"role": "assistant", 
+                                        "content": response, 
+                                        "feedback":feedback,
+                                        "verification_passed":None})
     export_outputs()
 
 def process_stata_query(qcat_result, uploaded_files):
@@ -655,22 +660,22 @@ def main():
 
     # Show KB summary on first visit (LLM-generated welcome summary)
     if not st.session_state.messages:
-        # if "kb_welcome_summary" not in st.session_state:
-        #     st.session_state.kb_welcome_summary = summarize_kb_for_welcome(cfg)
-        # kb_summary = st.session_state.kb_welcome_summary
-        welcome_msg = """
-        Welcome to WEAIBot, the AI-powered support assistant for the WEAI Helpdesk.\n\n
-        You can start by typing your question below, and WEAIBot will do its best to 
-        provide an answer based on available WEAI resources.\n\n
-        If you would like assistance from a member of the WEAI team, simply send a 
-        message through the chat and include the email address where we can reach you. 
-        A support ticket will be created, and a team member will contact you as soon as they are available. 
-        """
-        st.markdown(welcome_msg)
-
-        # if kb_summary:
-        #     with st.expander("Knowledge Base Contents", expanded=True):
-        #         st.markdown(kb_summary)
+        if "kb_welcome_summary" not in st.session_state:
+            st.session_state.kb_welcome_summary = summarize_kb_for_welcome(cfg)
+        kb_summary = st.session_state.kb_welcome_summary
+        # welcome_msg = """
+        # Welcome to WEAIBot, the AI-powered support assistant for the WEAI Helpdesk.\n\n
+        # You can start by typing your question below, and WEAIBot will do its best to 
+        # provide an answer based on available WEAI resources.\n\n
+        # If you would like assistance from a member of the WEAI team, simply send a 
+        # message through the chat and include the email address where we can reach you. 
+        # A support ticket will be created, and a team member will contact you as soon as they are available. 
+        # """
+        # st.markdown(welcome_msg)
+
+        if kb_summary:
+            with st.expander("Welcome!", expanded=True):
+                st.markdown(kb_summary)
 
     render_sidebar()
     render_chat()

src/prompts.py

@@ -22,8 +22,11 @@ ABSOLUTE RULE -- ZERO TOLERANCE FOR HALLUCINATION
   all. Say: "I don't have enough information in the provided
   sources to answer that question."
 - If the ONLY local source is the Knowledge Base Overview (a
-  summary of what files and datasets exist), use it to tell the
-  user what IS in the knowledge base, and note that the specific
+  summary of what files and datasets exist), or a requested 
+  resource cannot be found in WEAIbot's knowledge base, do not assume 
+  that it does not exist. Explain that the WEAI team may have access to 
+  additional resources that are not publicly available, and recommend 
+  contacting the WEAI Helpdesk for assistance. Also note that the specific
   information they asked about is not available in the current
   knowledge base.
 - Do NOT speculate, extrapolate, or fill in gaps with outside
@@ -46,6 +49,8 @@ CITATION RULES
    and attach the endnote number immediately after.
 3. Paraphrased claims still require a numbered endnote.
 4. Only use endnotes of references included in the reference list.
+5. ONLY use titles, authors, publication dates, and links (if available) 
+and DO NOT include any document paths
 
 ---------------------------------------------------------------------
 REFERENCE LIST FORMAT
@@ -149,6 +154,12 @@ RULES for SEARCH:
   in the knowledge base. This ensures better retrieval matching.
   (e.g., if the KB has a column "pts_a" for Political Terror Scale, rewrite
   "terror scores" as "pts_a Political Terror Scale scores")
+- ALWAYS include a sql query whenever a schema is provided and there are relevant documents in the sql database
+- Give detailed reasoning whenever you choose not to include a sql query.
+- DO NOT treat sources in the knowledge base as separate. Retrieve ALL relevant information from all available sources and present a unified response with the relevant citations.
+- ONLY use titles, authors, publication dates, and links (if available) and DO NOT include any document paths
+- ALWAYS note the specific WEAI version(s) in the user query and ONLY return results relevant to the version(s) specified. Ask for clarification if not sure
+
 
 RULES for CLARIFY:
 - ONLY ask for clarification when the question is genuinely ambiguous
@@ -163,7 +174,10 @@ RULES for CLARIFY:
   (e.g., instead of "Could you be more specific?", ask "Do you mean the
   PTS scores from the hr_data.csv dataset, or the CIRI scores from
   ciri_dataset.dta?")
-
+- ONLY use titles, authors, publication dates, and links (if available) and DO NOT include any document paths
+- DO NOT treat sources in the knowledge base as separate. Retrieve ALL relevant information from all available sources and present a unified response with the relevant citations.
+- ALWAYS note the specific WEAI version(s) in the user query and ONLY return results relevant to the version(s) specified. Ask for clarification if not sure
+  
 {sql_routing_block}
 {kb_overview_block}
 CONVERSATION HISTORY:
@@ -174,10 +188,10 @@ USER'S QUESTION: {query}
 Respond in JSON:
 {{
   "action": "search" or "clarify",
-  "route": "vector" or "sql" or "both",
+  "route": "both",
   "search_query": "keyword-optimized query for vector search (only if action is search)",
   "display_query": "clear natural-language question for the AI to answer (only if action is search)",
-  "sql_query": "SQLite SELECT query (only when route is sql or both)",
+  "sql_query": "SQLite SELECT query",
   "clarification_question": "question to ask (only if action is clarify)",
   "reasoning": "one sentence explaining your choice"
 }}
@@ -185,30 +199,10 @@ Respond in JSON:
 
 
 _SQL_ROUTING_INSTRUCTIONS = """\
-SQL ROUTING:
-You have access to structured SQL tables in addition to vector search.
-When choosing a route:
-- "sql" — for data lookups, filtering, aggregation, counting, comparisons
-  (e.g., "PTS scores for China", "how many countries", "average GDP")
-- "vector" — for conceptual, definitional, or methodology questions
-  (e.g., "what does PTS measure?", "explain the methodology")
-- "vector" — for meta-questions about what data is available
-  (e.g., "what datasets do you have?", "what is in the knowledge base?")
-  These are answered by codebooks and documentation, not by SQL queries.
-- "both" — for mixed questions combining concepts with data
-  (e.g., "explain PTS methodology and show China's scores")
-
-IMPORTANT: If the user asks about data summarization, counting, averaging,
-filtering, or any question that requires looking at dataset rows, ALWAYS
-set route to "sql" or "both".
-
-When route is "sql" or "both", you MUST provide a valid SQLite SELECT query
-in the "sql_query" field. When route is "vector", you SHOULD still include
-a "sql_query" if the data tables might answer the question as a fallback.
-Use exact equality (=) for text columns. The system will automatically
+You have access to structured SQL tables in addition to vector search.Provide a valid SQLite SELECT query
+in the "sql_query" field. Use exact equality (=) for text columns. The system will automatically
 retry with fuzzy matching (LIKE) if the exact query returns no results.
 Use only table/column names from the schema below.
-
 {schema_text}
 """
 

src/query_engine.py

@@ -112,18 +112,21 @@ def understand_query(
     from src.kb_meta import load_kb_meta
     kb_overview = load_kb_meta(cfg)
 
+    # print(user_query)
+    # print("\n")
     prompt = build_query_understanding_prompt(
         user_query, domain, history,
         sql_schema_summary=schema_summary,
         kb_overview=kb_overview,
     )
-
+    # print(prompt)
+    # print("\n")
     try:
         raw = generate(
             "You are a query reformulation assistant. Return only JSON.",
             prompt,
             cfg,
-            max_tokens=512,
+            max_tokens=1024,
         )
         result = _parse_qu_result(raw, user_query)
     except Exception:
@@ -136,7 +139,7 @@ def understand_query(
             "route": "vector",
             "sql_query": None,
         }
-
+    print(raw)
     return result
 
 def _parse_qcat_result(raw:str) -> Dict: