Update main.py

main.py

@@ -16,8 +16,9 @@ from services.kb_creation import (
     collection,
     ingest_documents,
     hybrid_search_knowledge_base,
-    get_section_text,             # NEW
-    get_best_steps_section_text,  # NEW
+    detect_user_intent,              # NEW semantic intent
+    get_section_text,
+    get_best_steps_section_text,
 )
 
 from services.login import router as login_router
@@ -92,6 +93,7 @@ GEMINI_URL = (
     f"gemini-2.5-flash-lite:generateContent?key={GEMINI_API_KEY}"
 )
 
+# ---------- Helpers: context merge + sanitation ----------
 def extract_kb_context(kb_results: Optional[Dict[str, Any]], top_chunks: int = 2) -> Dict[str, Any]:
     if not kb_results or not isinstance(kb_results, dict):
         return {"context": "", "sources": [], "top_hits": [], "context_found": False, "best_score": None, "best_combined": None}
@@ -162,6 +164,7 @@ def _build_clarifying_message() -> str:
         "Reply with these details and I’ll search again."
     )
 
+# ---------- Intent helpers ----------
 def _build_tracking_descriptions(issue_text: str, resolved_text: str) -> Tuple[str, str]:
     issue = (issue_text or "").strip()
     resolved = (resolved_text or "").strip()
@@ -292,7 +295,7 @@ def _filter_context_for_query(context: str, query: str) -> Tuple[str, Dict[str,
     kept = sentences[:MAX_SENTENCES_CONCISE]
     return "\n".join(kept).strip(), {'mode': 'concise', 'matched_count': 0, 'all_sentences': len(sentences)}
 
-# ---------- intent & action specific extractors ----------
+# ---------- intent & formatting extractors ----------
 STEP_LINE_REGEX = re.compile(r"^\s*(?:\d+[\.\)]\s+|[•\-]\s+)", re.IGNORECASE)
 NAV_LINE_REGEX  = re.compile(r"(navigate\s+to|>\s*)", re.IGNORECASE)
 
@@ -309,21 +312,6 @@ NON_PROC_PHRASES = [
 ]
 NON_PROC_ANY_REGEX = re.compile("|".join([re.escape(v) for v in NON_PROC_PHRASES]), re.IGNORECASE)
 
-ACTION_SYNS_FLAT = {
-    "create": ["create", "creation", "add", "new", "generate"],
-    "update": ["update", "modify", "change", "edit"],
-    "delete": ["delete", "remove"],
-    "navigate": ["navigate", "go to", "open"],
-}
-
-def _action_in_line(ln: str, target_actions: List[str]) -> bool:
-    s = (ln or "").lower()
-    for act in target_actions:
-        for syn in ACTION_SYNS_FLAT.get(act, [act]):
-            if syn in s:
-                return True
-    return False
-
 def _is_procedural_line(ln: str) -> bool:
     s = (ln or "").strip()
     if not s:
@@ -340,14 +328,11 @@ def _is_procedural_line(ln: str) -> bool:
         return True
     return False
 
-def _extract_steps_only(text: str, max_lines: Optional[int] = 12, target_actions: Optional[List[str]] = None) -> str:
+def _extract_steps_only(text: str, max_lines: Optional[int] = 12) -> str:
     lines = [ln.strip() for ln in (text or "").splitlines() if ln.strip()]
     kept = []
     for ln in lines:
         if _is_procedural_line(ln):
-            if target_actions:
-                if not _action_in_line(ln, target_actions):
-                    continue
             kept.append(ln)
             if max_lines is not None and len(kept) >= max_lines:
                 break
@@ -374,30 +359,39 @@ def _extract_errors_only(text: str, max_lines: int = 10) -> str:
     return "\n".join(kept).strip() if kept else (text or "").strip()
 
 def _format_steps_markdown(lines: List[str]) -> str:
-    """
-    Convert a list of step lines into a clean Markdown numbered list.
-    Keeps original order, trims whitespace, skips empty lines.
-    """
-    items = []
+    """Convert step lines to a clean Markdown numbered list."""
+    items: List[str] = []
     for i, ln in enumerate(lines, start=1):
         s = (ln or "").strip()
         if not s:
             continue
-        # If the line already has leading "1. " or "• ", strip it so numbering is consistent
+        # Strip existing numbering/bullets to avoid double-numbering
         s = re.sub(r"^\s*(?:\d+[\.\)]\s+|[•\-]\s+)", "", s).strip()
         items.append(f"{i}. {s}")
     return "\n".join(items).strip()
-   
+
+def _format_bullets_markdown(lines: List[str]) -> str:
+    items: List[str] = []
+    for ln in lines:
+        s = (ln or "").strip()
+        if not s:
+            continue
+        s = re.sub(r"^\s*(?:\d+[\.\)]\s+|[•\-]\s+)", "", s).strip()
+        items.append(f"- {s}")
+    return "\n".join(items).strip()
+
+# ---------- Health ----------
 @app.get("/")
 async def health_check():
     return {"status": "ok"}
 
+# ---------- Chat endpoint ----------
 @app.post("/chat")
 async def chat_with_ai(input_data: ChatInput):
     try:
         msg_norm = (input_data.user_message or "").lower().strip()
 
-        # --- Yes/No handlers ---
+        # Yes/No handlers
         if msg_norm in ("yes", "y", "sure", "ok", "okay"):
             return {
                 "bot_response": ("Great! Tell me what you’d like to do next — check another ticket, create an incident, or describe your issue."),
@@ -416,7 +410,7 @@ async def chat_with_ai(input_data: ChatInput):
                 "debug": {"intent": "end_conversation"},
             }
 
-        # --- Resolution ack ---
+        # Resolution acknowledgement
         is_llm_resolved = _classify_resolution_llm(input_data.user_message)
         if _has_negation_resolved(msg_norm):
             is_llm_resolved = False
@@ -469,7 +463,7 @@ async def chat_with_ai(input_data: ChatInput):
                     "debug": {"intent": "resolved_ack", "exception": True},
                 }
 
-        # --- Incident intent ---
+        # Incident intent
         if _is_incident_intent(msg_norm):
             return {
                 "bot_response": (
@@ -488,7 +482,7 @@ async def chat_with_ai(input_data: ChatInput):
                 "debug": {"intent": "create_ticket"},
             }
 
-        # --- Generic opener ---
+        # Generic opener → ask for details first
         if _is_generic_issue(msg_norm):
             return {
                 "bot_response": (
@@ -509,7 +503,7 @@ async def chat_with_ai(input_data: ChatInput):
                 "debug": {"intent": "generic_issue"},
             }
 
-        # --- Status intent ---
+        # Ticket status
         status_intent = _parse_ticket_status_intent(msg_norm)
         if status_intent:
             if status_intent.get("ask_number"):
@@ -560,39 +554,58 @@ async def chat_with_ai(input_data: ChatInput):
             except Exception as e:
                 raise HTTPException(status_code=500, detail=safe_str(e))
 
-        # --- Hybrid KB search ---
+        # ---- Hybrid KB search (semantic intent aware) ----
         kb_results = hybrid_search_knowledge_base(input_data.user_message, top_k=10, alpha=0.6, beta=0.4)
         kb_ctx = extract_kb_context(kb_results, top_chunks=2)
         context_raw = kb_ctx.get("context", "") or ""
+        detected_intent, detected_intent_conf = detect_user_intent(input_data.user_message)
 
         filtered_text, filt_info = _filter_context_for_query(context_raw, input_data.user_message)
         context = filtered_text
         context_found = bool(kb_ctx.get("context_found", False)) and bool(context.strip())
         best_distance = kb_ctx.get("best_score")
         best_combined = kb_ctx.get("best_combined")
-        detected_intent = kb_results.get("user_intent", "neutral")
-        actions = kb_results.get("actions", [])
         best_doc = kb_results.get("best_doc")
         top_meta = (kb_results.get("metadatas") or [{}])[0] if (kb_results.get("metadatas") or []) else {}
 
-        # --- FULL SECTION when strongly found & steps intent ---
-        if detected_intent == "steps" and best_doc:
-            # prefer full 'Process Steps' section from the best SOP
+        # ---- FULL SECTION when strongly found & steps intent ----
+        high_conf = (best_combined is not None and best_combined >= 0.70) and (detected_intent_conf >= 0.55)
+        if detected_intent == "steps" and best_doc and high_conf:
             full_steps = get_best_steps_section_text(best_doc)
-            if not full_steps:
-                # fallback: full text of the top section
-                sec = (top_meta or {}).get("section")
-                if sec:
-                    full_steps = get_section_text(best_doc, sec)
+            if not full_steps and top_meta.get("section"):
+                full_steps = get_section_text(best_doc, top_meta.get("section"))
             if full_steps:
-                # If confidence is high, return ALL procedural lines (no truncation)
-                high_conf = (best_combined is not None and best_combined >= 0.75)
-                context = _extract_steps_only(full_steps, max_lines=None if high_conf else 20, target_actions=actions)
+                # show all procedural lines (no truncation)
+                context = _extract_steps_only(full_steps, max_lines=None)
+
+        # ---- Permission/Errors/Prereqs → tips + escalation if available ----
+        if detected_intent in ("permission", "errors", "prereqs") and best_doc:
+            errors = get_section_text(best_doc, "Common Errors & Resolution")
+            escalation = get_section_text(best_doc, "Escalation Path")
+            resp_lines: List[str] = []
+            if errors:
+                resp_lines.append("**Resolution Tips:**")
+                resp_lines.extend([f"- {ln.strip()}" for ln in errors.splitlines() if ln.strip()])
+            if escalation:
+                resp_lines.append("\n**Escalation Path:**")
+                resp_lines.append(escalation.strip())
+            if resp_lines:
+                return {
+                    "bot_response": "\n".join(resp_lines),
+                    "status": "PARTIAL",
+                    "context_found": True,
+                    "ask_resolved": False,
+                    "suggest_incident": True,
+                    "followup": "Shall I create a ticket for WMS Support?",
+                    "top_hits": [],
+                    "sources": [],
+                    "debug": {"intent": detected_intent, "best_doc": best_doc},
+                }
 
-        # Intent-shaping (secondary; if not already handled above)
+        # Intent-shaped extraction (secondary)
         q = (input_data.user_message or "").lower()
         if detected_intent == "steps" or any(k in q for k in ["steps", "procedure", "perform", "do", "process"]):
-            context = _extract_steps_only(context, max_lines=None if (best_combined and best_combined >= 0.75) else 12, target_actions=actions)
+            context = _extract_steps_only(context, max_lines=None if (best_combined and best_combined >= 0.70) else 12)
         elif detected_intent == "errors" or any(k in q for k in ["error", "issue", "fail", "not working", "resolution", "fix"]):
             context = _extract_errors_only(context, max_lines=10)
         elif any(k in q for k in ["navigate", "navigation", "menu", "screen"]):
@@ -626,12 +639,11 @@ async def chat_with_ai(input_data: ChatInput):
                 "debug": {"used_chunks": 0, "second_try": second_try, "best_distance": best_distance, "best_combined": best_combined},
             }
 
-        # LLM rewrite (kept, but we still fallback cleanly)
+        # LLM rewrite (kept) — will be formatted if empty/fallback
         enhanced_prompt = (
             "From the provided context, output only the actionable steps/procedure relevant to the user's question. "
             "Use ONLY the provided context; do NOT add information that is not present. "
-            + ("Return ONLY lines containing the requested action verbs. " if actions else "")
-            + "Do NOT include document names, section titles, or 'Source:' lines.\n\n"
+            "Do NOT include document names, section titles, or 'Source:' lines.\n\n"
             f"### Context\n{context}\n\n"
             f"### Question\n{input_data.user_message}\n\n"
             "### Output\n"
@@ -658,33 +670,26 @@ async def chat_with_ai(input_data: ChatInput):
         if not bot_text.strip():
             bot_text = context
         bot_text = _strip_any_source_lines(bot_text).strip()
-        
-# If the intent is steps, render lines as a numbered Markdown list
-        if kb_results.get("user_intent", "neutral") == "steps":
-           raw_lines = [ln.strip() for ln in bot_text.splitlines() if ln.strip()]
-
-    # If everything is on a single line, split defensively on ". "
-           if len(raw_lines) == 1:
-              parts = [p.strip() for p in re.split(r"\.\s+(?=[A-Z0-9])", raw_lines[0]) if p.strip()]
-              raw_lines = parts if len(parts) > 1 else raw_lines
-
-    # 🔴 NEW: merge number-only lines with the next line
-        merged: list[str] = []
-        i = 0
-        while i < len(raw_lines):
-           curr = raw_lines[i]
-        # A number-only line (e.g., "1", "2", "3")
-           if re.fullmatch(r"\d+", curr) and (i + 1) < len(raw_lines):
-              nxt = raw_lines[i + 1].strip()
-            # Combine into one line: "1. <next line text>"
-              merged.append(f"{curr}. {nxt}")
-              i += 2  # skip the next line; already merged
-           else:
-               merged.append(curr)
-               i += 1
-
-    # Finally: normalize and render as Markdown numbered list
-        bot_text = _format_steps_markdown(merged)
+
+        # --- Steps Markdown formatting (merge numeric-only lines) ---
+        if detected_intent == "steps":
+            raw_lines = [ln.strip() for ln in bot_text.splitlines() if ln.strip()]
+            if len(raw_lines) == 1:
+                parts = [p.strip() for p in re.split(r"\.\s+(?=[A-Z0-9])", raw_lines[0]) if p.strip()]
+                raw_lines = parts if len(parts) > 1 else raw_lines
+
+            merged: List[str] = []
+            i = 0
+            while i < len(raw_lines):
+                curr = raw_lines[i]
+                if re.fullmatch(r"\d+[\.\)]?", curr) and (i + 1) < len(raw_lines):
+                    num = re.match(r"(\d+)", curr).group(1)
+                    merged.append(f"{num}. {raw_lines[i+1].strip()}")
+                    i += 2
+                else:
+                    merged.append(curr)
+                    i += 1
+            bot_text = _format_steps_markdown(merged)
 
         status = "OK" if (
             (best_combined is not None and best_combined >= gate_combined_ok)
@@ -712,7 +717,7 @@ async def chat_with_ai(input_data: ChatInput):
                 "filter_mode": filt_info.get("mode"),
                 "matched_count": filt_info.get("matched_count"),
                 "user_intent": detected_intent,
-                "actions": actions,
+                "user_intent_conf": detected_intent_conf,
                 "best_doc": best_doc,
             },
         }
@@ -722,6 +727,7 @@ async def chat_with_ai(input_data: ChatInput):
     except Exception as e:
         raise HTTPException(status_code=500, detail=safe_str(e))
 
+# ---------- Incident endpoints ----------
 def _set_incident_resolved(sys_id: str) -> bool:
     try:
         token = get_valid_token()
@@ -842,7 +848,7 @@ async def generate_ticket_desc_ep(input_data: TicketDescInput):
         )
         headers = {"Content-Type": "application/json"}
         payload = {"contents": [{"parts": [{"text": prompt}]}]}
-        resp = requests.post(GEMINI_URL, headers=headers, json=payload, timeout=25, verify=GEMINI_GSL_VERIFY if 'GEMINI_GSL_VERIFY' in globals() else GEMINI_SSL_VERIFY)
+        resp = requests.post(GEMINI_URL, headers=headers, json=payload, timeout=25, verify=GEMINI_SSL_VERIFY)
         try:
             data = resp.json()
         except Exception:
@@ -904,7 +910,7 @@ async def incident_status(input_data: TicketStatusInput):
     except Exception as e:
         raise HTTPException(status_code=500, detail=safe_str(e))
 
-# ---- Admin endpoints (optional) ----
+# ---- Admin endpoints ----
 @app.get("/kb/info")
 async def kb_info():
     from services.kb_creation import get_kb_runtime_info