handling headline realted questions

components/handlers/whatsapp_handlers.py

@@ -1,7 +1,17 @@
 # handlers/whatsapp_handlers.py
 import logging
+import re
+from typing import Optional, Dict
+
 from fastapi.responses import JSONResponse
+
 from components.gateways.headlines_to_wa import fetch_cached_headlines, send_to_whatsapp
+from components.indexers.news_indexer import load_news_index  # should return a LlamaIndex VectorStoreIndex
+from components.LLMs.Mistral import MistralTogetherClient, build_messages
+
+# ------------------------------------------------------------
+# Utilities
+# ------------------------------------------------------------
 
 def _safe_send(text: str, to: str) -> dict:
     """Wrap send_to_whatsapp with logging & safe error handling."""
@@ -17,6 +27,10 @@ def _safe_send(text: str, to: str) -> dict:
         return {"status": "error", "error": str(e)}
 
 
+# ------------------------------------------------------------
+# Headlines
+# ------------------------------------------------------------
+
 def handle_headlines(from_number: str) -> JSONResponse:
     full_message_text = fetch_cached_headlines()
 
@@ -36,6 +50,10 @@ def handle_headlines(from_number: str) -> JSONResponse:
         return JSONResponse(status_code=500, content={"status": "error", "message": "Failed to send digest"})
 
 
+# ------------------------------------------------------------
+# Preferences / Greeting / Help / Unsubscribe / Small Talk
+# ------------------------------------------------------------
+
 def handle_preferences(from_number: str) -> JSONResponse:
     msg = (
         "Let’s tune your feed. Reply with topics you like:\n"
@@ -79,12 +97,145 @@ def handle_small_talk(from_number: str) -> JSONResponse:
     return JSONResponse(status_code=200, content={"status": "success", "message": "Small talk"})
 
 
+# ------------------------------------------------------------
+# Chat Question → “Explain by number” flow
+# ------------------------------------------------------------
+
+_HEADLINE_LINE_RE = re.compile(r"^\s*(\d+)\.\s+(.*)$")
+
+def _extract_number_ref(text: str) -> Optional[int]:
+    """
+    Find a referenced headline number in free text, e.g.:
+    'explain number 14', 'no. 7 please', '#9', '14', 'explain 14 like I am 5'
+    Returns int or None.
+    """
+    s = (text or "").lower()
+
+    # explicit forms
+    m = re.search(r"(?:number|no\.?|num|#)\s*(\d+)", s)
+    if m:
+        return int(m.group(1))
+
+    # a bare number (avoid picking up years like 2025; cap at 1..200)
+    m2 = re.search(r"\b(\d{1,3})\b", s)
+    if m2:
+        n = int(m2.group(1))
+        if 1 <= n <= 200:
+            return n
+
+    return None
+
+
+def _parse_rendered_digest(rendered: str) -> Dict[int, str]:
+    """
+    Parse the same rendered digest string you send on WhatsApp and build a map:
+    { number -> headline_line_text }
+    """
+    mapping: Dict[int, str] = {}
+    for line in (rendered or "").splitlines():
+        m = _HEADLINE_LINE_RE.match(line)
+        if not m:
+            continue
+        num = int(m.group(1))
+        headline_txt = m.group(2).strip()
+        mapping[num] = headline_txt
+    return mapping
+
+
+def _retrieve_context_for_headline(headline_text: str, top_k: int = 5) -> str:
+    """
+    Use the vector index to pull contextual passages related to the headline.
+    Gracefully degrades if index is unavailable.
+    """
+    try:
+        index = load_news_index()
+        try:
+            qe = index.as_query_engine(similarity_top_k=top_k)
+        except Exception:
+            # Older LlamaIndex fallback
+            from llama_index.core.query_engine import RetrievalQueryEngine
+            qe = RetrievalQueryEngine(index=index, similarity_top_k=top_k)
+
+        query = (
+            "Retrieve concise, factual context that best explains this headline:\n"
+            f"{headline_text}\n"
+            "Focus on who/what/when/where/why, include crucial numbers, avoid speculation."
+        )
+        resp = qe.query(query)
+        return str(resp)
+    except Exception as e:
+        logging.exception(f"Vector retrieval failed: {e}")
+        return ""
+
+
+def _eli5_answer(question: str, context: str) -> str:
+    """
+    Ask Mistral (via Together) to explain using simple words.
+    """
+    sys_prompt = (
+        "You are a concise explainer for a news assistant. "
+        "Answer like the person is 5 years old (ELI5): short sentences, simple words, "
+        "and 3–6 bullet points max. Be accurate and neutral. If unsure, say so."
+    )
+    user_prompt = (
+        f"QUESTION:\n{question}\n\n"
+        f"CONTEXT (may be partial):\n{context}\n\n"
+        "Now give a short ELI5 explanation. Avoid jargon. If numbers matter, include them."
+    )
+
+    try:
+        llm = MistralTogetherClient()
+        msgs = build_messages(user_prompt, sys_prompt)
+        out, _usage = llm.chat(msgs, temperature=0.2, max_tokens=350)
+        return out.strip()
+    except Exception as e:
+        logging.exception(f"Mistral ELI5 generation failed: {e}")
+        return "I couldn’t generate a simple explanation right now. Please try again."
+
+
 def handle_chat_question(from_number: str, message_text: str) -> JSONResponse:
-    # Placeholder: integrate with /ask later
+    """
+    Smart handler:
+      - If the user references a headline number (“explain 14 like I’m 5”),
+        1) Parse the number
+        2) Look up that numbered line from the rendered digest
+        3) Retrieve vector context
+        4) Generate an ELI5 answer with Mistral (Together.ai)
+      - Otherwise, provide a gentle hint (for now).
+    """
+    logging.info(f"Chat question from {from_number}: {message_text}")
+
+    # 1) Try to find a headline number reference
+    number = _extract_number_ref(message_text or "")
+    if number is not None:
+        # 2) Load rendered digest and map numbers to lines
+        rendered = fetch_cached_headlines()
+        mapping = _parse_rendered_digest(rendered)
+        target_line = mapping.get(number)
+
+        if not target_line:
+            _safe_send(
+                f"I couldn’t find headline *{number}* in today’s digest. "
+                "Try another number or say *headlines* to see today’s list.",
+                to=from_number,
+            )
+            return JSONResponse(status_code=200, content={"status": "success", "message": "Number not found"})
+
+        # 3) Retrieve context from the vector index using the headline line
+        ctx = _retrieve_context_for_headline(target_line, top_k=5)
+
+        # 4) Generate ELI5 answer
+        question = f"Explain headline #{number}: {target_line}"
+        answer = _eli5_answer(question, ctx)
+
+        # 5) Send back
+        _safe_send(answer, to=from_number)
+        return JSONResponse(status_code=200, content={"status": "success", "message": "ELI5 sent"})
+
+    # No number found → for now, guide the user
     _safe_send(
-        "Great question! I’m setting up deep Q&A. For now, type *headlines* for today’s digest "
-        "or *preferences* to tune your feed.",
+        "Ask me about a specific headline by number, e.g., *explain 7 like I’m 5*.\n"
+        "Or type *headlines* for today’s digest.",
         to=from_number,
     )
-    logging.info(f"Chat question from {from_number}: {message_text}")
-    return JSONResponse(status_code=200, content={"status": "success", "message": "Question acknowledged"})
+    return JSONResponse(status_code=200, content={"status": "success", "message": "Generic reply"})