Update src/agent.py

src/agent.py

@@ -58,4 +58,377 @@ PURE_GREETING_RE = re.compile(r"^(hi|hello|hey|good morning|good afternoon|good
 GREETING_FOLLOWUP_RE = re.compile(r"^(yes|yes please|yes check|sure|okay|ok|yeah|yep|please do|go ahead)$", re.IGNORECASE)
 HOURS_RE = re.compile(r"\b(library hours?|opening hours?|closing hours?|open hours?|open today|open tomorrow|open now|open on|open this|closed today|closed tomorrow|closed on|closed this|when.*open|when.*close|when.*does.*library|what time.*open|what time.*close|what time.*library|opening time|closing time|is.*library.*open|is.*campus.*library|is.*habshan.*open|is.*san.*open|is.*library.*closed|will.*library.*open|will.*it.*open|does.*library.*open|library.*open.*today|library.*close.*today|habshan.*open|san.*open|main campus.*open|ramadan hours?|weekend hours?|friday hours?|saturday hours?|sunday hours?|library.*schedule|schedule.*library|what.*hours|till.*when|until.*when|what day.*open|which day.*open|what day.*closed|which day.*closed)\b", re.IGNORECASE)
 KU_CAMPUS_RE = re.compile(r"\b(where is the library|where is.*library|library.*location|location.*library|which building|what building|which floor|main campus library|san campus library|habshan library|sas al nakhl library|library.*address|address.*library|how to get to.*library|how.*reach.*library|how.*find.*library|library.*campus(es)?|campus(es)?.*library|how many.*librar|how many.*branch|library.*branch(es)?|branch(es)?.*library|which campus.*library|library.*which campus|library.*map|map.*library|how many libraries|what libraries|which libraries|(ku|khalifa).*(has|have|got).*(librar|branch)|(librar|branch).*(ku|khalifa).*(has|have)|(all|both|two|2).*(libraries|library|branches)|tell me about.*(campus|libraries|branches)|describe.*(campus|libraries)|(what|where).*(campus|campuses).*(librar))\b", re.IGNORECASE)
-RESEARCH_CUE_RE = re.compile(r"\b(find|search|look for|show me|get me|give me|locate|recommend|suggest|articles?|papers?|books?|journals?|studies|literature|research|systematic review|evidence|sources|database|databases|peer reviewed|open access|advances?|approach(es)?|method(s|ology)?|*
+RESEARCH_CUE_RE = re.compile(r"\b(find|search|look for|show me|get me|give me|locate|recommend|suggest|articles?|papers?|books?|journals?|studies|literature|research|systematic review|evidence|sources|database|databases|peer reviewed|open access|advances?|approach(es)?|method(s|ology)?|technique[s]?|application[s]?|framework[s]?|model[s]?|algorithm[s]?|publication[s]?|recent work|current state|state of the art)\b", re.IGNORECASE)
+LIBRARY_CUE_RE = re.compile(r"\b(library|librarian|borrow|loan|renew|fine|study room|room booking|reserve a room|account|my library|interlibrary|ill|khazna|orcid|open access|apc|refworks|libkey|hours|location|contact|visitor|alumni|database access|off campus|remote access)\b", re.IGNORECASE)
+MEDICAL_SEARCH_RE = re.compile(r"\b(pubmed|embase|cinahl|clinicalkey|cochrane|uptodate|medline|systematic review|clinical trial|biomedical literature|medical literature)\b", re.IGNORECASE)
+MEDICAL_KEYWORDS = [
+    'medicine','medical','health','clinical','nursing','pharmacy','biomedical',
+    'anatomy','physiology','epidemiology','oncology','cardiology','neuroscience',
+    'genomics','pathology','surgery','mental health','nutrition','public health',
+    'disease','therapy','treatment','diagnosis','patient','drug','pharmaceutical',
+    'hospital','cancer','diabetes','heart','brain','cell biology','immunology',
+    'biochemistry','microbiology','virology','pediatric','psychiatric','dental',
+    'ophthalmology','radiology','anesthesia','emergency medicine','geriatric',
+    'crispr','gene therapy','gene editing','genomic medicine','precision medicine',
+]
+
+CURRENT_INFO_RE = re.compile(r"\b(current|currently|latest|today|recent|recently|now|this week|this month|news|update|updated|president|prime minister|ceo|minister|king|ruler|who won)\b", re.IGNORECASE)
+
+# Hardcoded answers
+LIBRARY_HOURS_URL = "https://library.ku.ac.ae/hours"
+KU_MAIN_URL = "https://www.ku.ac.ae/"
+
+GROUNDED_LIBRARY_MAP = {
+    "ill": "interlibrary loan ILL document delivery request article request book unavailable not in collection borrow from another library",
+    "fulltext": "full text libkey nomad article access pdf article not available no full text get full text request article",
+    "primo": "PRIMO discovery catalog holdings publication finder library has this journal books articles search catalog",
+    "circulation": "borrowing circulation renew renewal loan period due date hold request fines fees my library account checked out overdue",
+    "research_help": "Ask a Librarian research consultation reference help subject guides research skills instruction tours workshops",
+    "orcid_oa": "ORCID open access APC publishing research impact Scopus SciVal bibliometrics openaccess",
+    "database_access": "database access e-resources remote access off campus login vendor issue Rani Anand e-books patents",
+    "medical_help": "medical librarian Jason Fetty PubMed Embase CINAHL Cochrane UpToDate systematic review clinical databases",
+    "systems_help": "Walter Brian Hall systems website technology digital services library systems coding STEAM open access APC",
+    "acquisitions": "Alia Al-Harrasi Meera Alnaqbi acquisitions request title suggest a book collection development technical services cataloguing metadata",
+    "cataloguing": "Shamsa Alhosani William Hutchinson Alia Al-Harrasi cataloguing cataloging metadata technical services",
+    "rooms": "study room reserve a room room booking group study rooms libcal book room room reservation",
+}
+
+# LLM Classifier Prompt
+_CLASSIFIER_SYSTEM = """You are the routing brain of LibBee, the Khalifa University Library AI Assistant (Abu Dhabi, UAE).
+Your only job: read the user message and return a single JSON object routing it to the right handler.
+
+INTENTS — pick exactly one:
+  "social"         — casual chat, greetings, emotions, compliments, complaints, vague help requests
+  "library_info"   — questions about KU Library services, hours, rooms, borrowing, fines, ILL, staff, databases, ORCID, RefWorks, open access, APC, account, printing, remote access, LibKey, acquisitions, Khazna, thesis submission, library policies, alumni access
+  "search_academic"— user wants to FIND articles / papers / books / literature on a specific academic topic
+  "search_medical" — same as search_academic but topic is medical/clinical/nursing/pharmacy/health sciences
+  "general_recent" — questions about current events, news, latest info, who holds a position now, real-time data (stock prices, today's weather, breaking news)
+  "general"        — factual/knowledge question NOT about KU Library and NOT a research search (explain a concept, historical facts, definitions)
+
+RULES:
+  1. When in doubt between library_info and social, choose library_info if ANY library keyword is present.
+  2. "I don't know how to search for articles" = library_info (has library task content).
+  3. "I am fed up with searching" alone (no specific topic/database) = social (expressing frustration).
+  4. "I need help finding articles on diabetes" = search_medical.
+  5. KU = Khalifa University. Do NOT confuse with Kuwait University.
+  6. ANY question asking about advances, approaches, methods, effectiveness, impact, shifts, trends, bottlenecks, or relationship on a specific academic/scientific topic = search_academic or search_medical.
+
+For "social" intent ONLY: also include "casual_answer" — a warm 1-3 sentence LibBee response (friendly librarian tone, no markdown, no bullet points). Offer to help with library services.
+For all other intents: omit "casual_answer" entirely.
+
+Respond ONLY with valid JSON, no preamble, no markdown fences.
+
+Examples:
+{"intent":"social","casual_answer":"Of course, I'm here to help! I'm LibBee, the KU Library AI Assistant. What would you like to find today?"}
+{"intent":"library_info"}
+{"intent":"search_academic"}
+{"intent":"search_medical"}
+{"intent":"general_recent"}
+{"intent":"general"}"""
+
+# ========== HELPER FUNCTIONS ==========
+
+def _is_greeting_menu_followup(question: str, history: List[ChatMessage]) -> bool:
+    q = (question or "").strip()
+    if not GREETING_FOLLOWUP_RE.match(q):
+        return False
+    # Check last assistant message
+    for m in reversed(history):
+        if m.role == "assistant":
+            last = m.content.lower()
+            return "i'm <strong>libbee</strong>" in last or "are you looking for one of these" in last
+    return False
+
+def _greeting_menu_clarify_answer() -> str:
+    return (
+        "Sure — what would you like help with?<br><br>"
+        "You can choose one of these or type your question directly:<br>"
+        "• Find articles and books<br>"
+        "• Get full text for an article<br>"
+        "• Submit an Interlibrary Loan (ILL) request<br>"
+        "• Contact a librarian<br>"
+        "• Check library hours"
+    )
+
+def _looks_library_hours_question(question: str) -> bool:
+    q = question.strip().lower()
+    return bool(HOURS_RE.search(q)) and any(
+        kw in q for kw in (
+            'library', 'campus', 'habshan', 'san', 'hours', 'open', 'close',
+            'schedule', 'till', 'until', 'today', 'tomorrow', 'friday',
+            'saturday', 'sunday', 'week', 'ramadan'
+        )
+    )
+
+def _looks_campus_question(question: str) -> bool:
+    return bool(KU_CAMPUS_RE.search(question))
+
+def _looks_medical_search(question: str) -> bool:
+    q = question.lower()
+    if not any(kw in q for kw in MEDICAL_KEYWORDS):
+        return False
+    return bool(MEDICAL_SEARCH_RE.search(q) or RESEARCH_CUE_RE.search(q))
+
+def _looks_research_question(question: str) -> bool:
+    q = question.lower()
+    if RESEARCH_CUE_RE.search(q):
+        return True
+    return False
+
+def _hours_redirect_answer() -> str:
+    return (
+        "Here are the current <strong>Khalifa University Library</strong> hours:<br><br>"
+        "<strong>Habshan Library (SAN Campus):</strong><br>"
+        "Sunday – Wednesday: 8:00 AM – 7:00 PM<br>"
+        "Thursday: 8:00 AM – 4:00 PM<br>"
+        "Friday: Closed<br>"
+        "Saturday: 12:00 PM – 8:00 PM<br><br>"
+        "⚠️ Hours may change during exam periods, public holidays, and Ramadan. "
+        "Always check the official Library Hours page for the latest times:<br>"
+        f'<a href="{LIBRARY_HOURS_URL}" target="_blank">{LIBRARY_HOURS_URL}</a>'
+    )
+
+def _ku_campus_answer() -> str:
+    return (
+        "Khalifa University Library has <strong>two branches</strong>:<br><br>"
+        "<strong>1. Main Campus Library</strong><br>"
+        "Location: E Building, Main Campus, Abu Dhabi<br>"
+        "Reference desk: <a href=\"tel:+97123124604\">+971 2 312 4604</a><br><br>"
+        "<strong>2. Habshan Library (SAN Campus)</strong><br>"
+        "Location: Building 5, Ground Floor &amp; First Floor, "
+        "Sas Al Nakhl (SAN) Campus, Abu Dhabi<br>"
+        "Service desk: <a href=\"tel:+97123123160\">+971 2 312 3160</a><br><br>"
+        "Both branches offer study spaces, computer workstations, printing, scanning, "
+        "and access to physical collections.<br><br>"
+        "For general library enquiries: "
+        "<a href=\"https://library.ku.ac.ae/AskUs\" target=\"_blank\">Ask a Librarian</a> "
+        "or email <a href=\"mailto:libse@ku.ac.ae\">libse@ku.ac.ae</a>."
+    )
+
+def _ku_general_redirect_answer() -> str:
+    return (
+        f"I'm <strong>LibBee</strong>, the Khalifa University <strong>Library</strong> AI Assistant, so I'm best for library resources, services, databases, research help, and staff contacts.<br><br>"
+        f"For general Khalifa University questions, please visit the main KU website: <a href=\"{KU_MAIN_URL}\" target=\"_blank\">{KU_MAIN_URL}</a>. "
+        "A broader university chatbot is available there for non-library questions."
+    )
+
+def _is_social_or_greeting(question: str) -> bool:
+    return bool(SOCIAL_RE.match((question or "").strip()))
+
+async def _llm_classify(question: str, history: List[ChatMessage], model: str) -> Dict:
+    """Classify intent using LLM"""
+    settings = get_settings()
+    
+    if not settings.openai_api_key:
+        # Fallback to rule-based if no API key
+        q = question.lower()
+        if SOCIAL_RE.match(q):
+            return {"intent": "social", "casual_answer": "Hello! I'm LibBee, the KU Library AI Assistant. I'm here to help you with articles, books, databases, and library services. What would you like to find today?"}
+        elif _looks_medical_search(question):
+            return {"intent": "search_medical"}
+        elif _looks_research_question(question):
+            return {"intent": "search_academic"}
+        elif LIBRARY_CUE_RE.search(q):
+            return {"intent": "library_info"}
+        elif CURRENT_INFO_RE.search(q):
+            return {"intent": "general_recent"}
+        else:
+            return {"intent": "general"}
+    
+    try:
+        ctx_parts = []
+        for m in history[-4:]:
+            if m.role in ("user", "assistant") and m.content:
+                ctx_parts.append(f"{m.role}: {m.content[:100]}")
+        ctx = " | ".join(ctx_parts) if ctx_parts else "none"
+
+        user_msg = f"Message: {question[:400]}\nContext: {ctx}\nJSON:"
+
+        if model == "claude" and settings.anthropic_api_key:
+            from langchain_anthropic import ChatAnthropic
+            llm = ChatAnthropic(model="claude-3-haiku-20240307", temperature=0, max_tokens=150)
+        else:
+            from langchain_openai import ChatOpenAI
+            llm = ChatOpenAI(model="gpt-4o-mini", temperature=0, max_tokens=150)
+
+        response = llm.invoke([
+            {"role": "system", "content": _CLASSIFIER_SYSTEM},
+            {"role": "user", "content": user_msg},
+        ])
+        
+        raw = response.content.strip()
+        if raw.startswith("```"):
+            raw = raw.split("\n", 1)[1].rsplit("```", 1)[0].strip()
+        
+        result = json.loads(raw)
+        intent = result.get("intent", "general")
+        if intent not in {"social", "library_info", "search_academic", "search_medical", "general_recent", "general"}:
+            intent = "general"
+        
+        return {"intent": intent, "casual_answer": result.get("casual_answer", "")}
+    
+    except Exception as e:
+        print(f"Classifier error: {e}")
+        # Fallback
+        return {"intent": "general", "casual_answer": ""}
+
+async def _libbee_casual_response(question: str, history: List[ChatMessage], model: str, hint: str = "") -> str:
+    """Generate casual response"""
+    if hint and len(hint.strip()) > 25:
+        return hint.strip()
+    
+    settings = get_settings()
+    behavior = "You are LibBee, the Khalifa University Library AI Assistant. Respond warmly and naturally in 1-3 sentences as a friendly, knowledgeable librarian."
+    
+    try:
+        if model == "claude" and settings.anthropic_api_key:
+            from langchain_anthropic import ChatAnthropic
+            llm = ChatAnthropic(model="claude-3-haiku-20240307", temperature=0.6, max_tokens=200)
+        else:
+            from langchain_openai import ChatOpenAI
+            llm = ChatOpenAI(model="gpt-4o-mini", temperature=0.6, max_tokens=200)
+        
+        msgs = [{"role": "system", "content": behavior}]
+        for m in history[-4:]:
+            if m.role in ("user", "assistant"):
+                msgs.append({"role": m.role, "content": m.content[:200]})
+        msgs.append({"role": "user", "content": question})
+        
+        return llm.invoke(msgs).content.strip()
+    except Exception:
+        return "I'm <strong>LibBee</strong>, the Khalifa University Library AI Assistant — always happy to help! What can I find for you today?"
+
+# ========== MAIN ENDPOINT ==========
+
+@router.post("", response_model=AgentResponse)
+async def agent_query(request: Request, req: AgentRequest):
+    start_time = time.time()
+    question = req.question.strip()
+    history = req.history
+    model = req.model
+    
+    # LAYER 0: Instant Handlers
+    
+    # 1. Pure greeting
+    if PURE_GREETING_RE.match(question):
+        return AgentResponse(
+            answer=(
+                "Hi! I'm <strong>LibBee</strong>, the Khalifa University Library AI Assistant.<br><br>"
+                "I can help you find articles and books, search databases, access full text, "
+                "request Interlibrary Loan (ILL), and answer questions about library services and staff.<br><br>"
+                "Are you looking for one of these? You can also type your question directly."
+            ),
+            intent="social_greeting",
+            tools_used=[],
+            model_used=model,
+            response_time=time.time() - start_time
+        )
+    
+    # 2. Staff name match
+    staff_match = match_staff_name(question)
+    if staff_match:
+        return AgentResponse(
+            answer=staff_name_answer(staff_match),
+            intent="library_info",
+            tools_used=["staff_name_match"],
+            model_used=model,
+            response_time=time.time() - start_time
+        )
+    
+    # 3. Staff role match
+    staff_role_match = match_staff_role(question)
+    if staff_role_match:
+        return AgentResponse(
+            answer=staff_role_answer(staff_role_match, question),
+            intent="library_info",
+            tools_used=["staff_role_match"],
+            model_used=model,
+            response_time=time.time() - start_time
+        )
+    
+    # 4. Greeting menu followup
+    if _is_greeting_menu_followup(question, history):
+        return AgentResponse(
+            answer=_greeting_menu_clarify_answer(),
+            intent="social_greeting",
+            tools_used=[],
+            model_used=model,
+            is_follow_up=True,
+            response_time=time.time() - start_time
+        )
+    
+    # 5. Library hours
+    if _looks_library_hours_question(question):
+        return AgentResponse(
+            answer=_hours_redirect_answer(),
+            intent="library_info",
+            tools_used=["hours_hardcoded"],
+            model_used=model,
+            ask_librarian=False,
+            response_time=time.time() - start_time
+        )
+    
+    # 6. Campus/location
+    if _looks_campus_question(question):
+        return AgentResponse(
+            answer=_ku_campus_answer(),
+            intent="library_info",
+            tools_used=["campus_hardcoded"],
+            model_used=model,
+            ask_librarian=False,
+            response_time=time.time() - start_time
+        )
+    
+    # LAYER 1: LLM Classification
+    cls_result = await _llm_classify(question, history, model)
+    intent = cls_result.get("intent", "general")
+    casual_hint = cls_result.get("casual_answer", "")
+    
+    # LAYER 2: Execute by Intent
+    
+    if intent == "social":
+        casual = await _libbee_casual_response(question, history, model, casual_hint)
+        return AgentResponse(
+            answer=casual,
+            intent="social_greeting",
+            tools_used=["libbee_casual"],
+            model_used=model,
+            response_time=time.time() - start_time
+        )
+    
+    elif intent == "library_info":
+        # For now, return helpful message. Later add RAG here.
+        return AgentResponse(
+            answer="I can help with library information. For specific questions about hours, locations, or services, please check the Library Services tab, or ask me about:<br>• Library hours<br>• Study room booking<br>• Interlibrary Loan (ILL)<br>• Database access",
+            intent="library_info",
+            tools_used=["get_library_info"],
+            model_used=model,
+            ask_librarian=True,
+            response_time=time.time() - start_time
+        )
+    
+    elif intent in ["search_academic", "search_medical"]:
+        # Return search guidance
+        return AgentResponse(
+            answer=f"I'll help you search for {'medical' if intent == 'search_medical' else 'academic'} literature on: <strong>{question}</strong>.<br><br>Use the <strong>AI Search</strong> tab for best results, or I can help you refine your search terms.",
+            intent=intent,
+            tools_used=["search_primo"],
+            model_used=model,
+            corrected_query=question,
+            natural_query=question,
+            database_query=question,
+            response_time=time.time() - start_time
+        )
+    
+    else:  # general or general_recent
+        return AgentResponse(
+            answer=_ku_general_redirect_answer(),
+            intent="general",
+            tools_used=["ku_general_redirect"],
+            model_used=model,
+            response_time=time.time() - start_time
+        )
+
+@router.get("/test")
+async def test():
+    return {"status": "ok", "message": "Agent endpoint is working!", "staff_loaded": len(STAFF_DIRECTORY)}