checkpoint 3

core/backend.py

@@ -11,6 +11,7 @@ from datetime import datetime, timedelta
 from dotenv import load_dotenv
 import re
 import traceback
+from difflib import SequenceMatcher
 
 from langchain_core.messages import (
     AIMessage, AIMessageChunk, HumanMessage, RemoveMessage,
@@ -208,15 +209,23 @@ TOOL_INTENT_WORDS = (
 )
 
 SPECIALTY_ALIASES = {
-    "চক্ষু": ["ophthalmologist", "eye specialist", "eye doctor", "ophthalmology", "eye"],
-    "আই": ["ophthalmologist", "eye specialist", "eye doctor", "ophthalmology", "eye"],
-    "চোখ": ["ophthalmologist", "eye specialist", "eye doctor", "ophthalmology", "eye"],
-    "হৃদরোগ": ["cardiologist", "heart", "cardio", "cardiology"],
-    "কার্ডিও": ["cardiologist", "heart", "cardio", "cardiology"],
+    "চক্ষু": ["eye specialist", "ophthalmologist", "ophthalmology", "eye doctor", "eye"],
+    "আই": ["eye specialist", "ophthalmologist", "ophthalmology", "eye doctor", "eye"],
+    "চোখ": ["eye specialist", "ophthalmologist", "ophthalmology", "eye doctor", "eye"],
+    "eye": ["eye specialist", "ophthalmologist", "ophthalmology", "eye doctor"],
+    "eye specialist": ["eye specialist", "ophthalmologist", "ophthalmology", "eye doctor", "eye"],
+    "ophthalmology": ["eye specialist", "ophthalmologist", "ophthalmology", "eye doctor", "eye"],
+    "হৃদরোগ": ["cardiologist", "cardiology", "heart specialist", "heart", "cardio"],
+    "কার্ডিও": ["cardiologist", "cardiology", "heart specialist", "heart", "cardio"],
+    "cardio": ["cardiologist", "cardiology", "heart specialist", "heart"],
+    "cardiology": ["cardiologist", "cardiology", "heart specialist", "heart"],
     "মেডিসিন": ["medicine", "internal medicine", "physician", "general medicine"],
-    "নিউরো": ["neurologist", "neurology", "brain"],
-    "স্নায়ু": ["neurologist", "neurology", "brain"],
-    "নিউরোলজি": ["neurologist", "neurology", "neorology", "neuro"],
+    "নিউরো": ["neurologist", "neurology", "neuro", "brain specialist", "brain"],
+    "স্নায়ু": ["neurologist", "neurology", "neuro", "brain specialist", "brain"],
+    "নিউরোলজি": ["neurologist", "neurology", "neorology", "neuro", "brain specialist", "brain"],
+    "neuro": ["neurologist", "neurology", "brain specialist", "brain"],
+    "neurology": ["neurologist", "neurology", "brain specialist", "brain"],
+    "neurologist": ["neurologist", "neurology", "brain specialist", "brain"],
     "নাক": ["ent", "otolaryngologist", "ear nose throat"],
     "কান": ["ent", "otolaryngologist", "ear nose throat"],
     "গলা": ["ent", "otolaryngologist", "ear nose throat"],
@@ -224,27 +233,49 @@ SPECIALTY_ALIASES = {
     "স্কিন": ["dermatologist", "skin", "dermatology"],
     "ডেন্টাল": ["dentist", "dental", "teeth"],
     "দাঁত": ["dentist", "dental", "teeth"],
+    "dentist": ["dentist", "dental", "teeth"],
     "গাইনী": ["gynecologist", "gynaecologist", "obgyn", "women"],
     "মহিলা": ["gynecologist", "gynaecologist", "obgyn", "women"],
-    "শিশু": ["pediatrician", "child", "children"],
-    "পেডিয়াট্রিক": ["pediatrician", "child", "children"],
-    "অর্থো": ["orthopedic", "orthopaedic", "bone"],
-    "হাড়": ["orthopedic", "orthopaedic", "bone"],
+    "শিশু": ["child specialist", "pediatrician", "pediatrics", "child doctor", "children"],
+    "পেডিয়াট্রিক": ["child specialist", "pediatrician", "pediatrics", "child doctor", "children"],
+    "child": ["child specialist", "pediatrician", "pediatrics", "child doctor", "children"],
+    "child specialist": ["child specialist", "pediatrician", "pediatrics", "child doctor", "children"],
+    "pediatrician": ["child specialist", "pediatrician", "pediatrics", "child doctor", "children"],
+    "pediatrics": ["child specialist", "pediatrician", "pediatrics", "child doctor", "children"],
+    "অর্থো": ["orthopedics", "orthopedic", "orthopaedic", "orthopaedics", "ortho", "bone"],
+    "হাড়": ["orthopedics", "orthopedic", "orthopaedic", "orthopaedics", "ortho", "bone"],
+    "orthopedics": ["orthopedics", "orthopedic", "orthopaedic", "orthopaedics", "ortho", "bone"],
+    "orthopedic": ["orthopedics", "orthopedic", "orthopaedic", "orthopaedics", "ortho", "bone"],
+    "orthopaedic": ["orthopedics", "orthopedic", "orthopaedic", "orthopaedics", "ortho", "bone"],
     "বক্ষ": ["chest", "pulmonologist", "respiratory"],
     "শ্বাস": ["pulmonologist", "respiratory", "chest"],
     "কিডনি": ["nephrologist", "kidney", "renal"],
-    "গ্যাস্ট্রো": ["gastroenterologist", "stomach", "digestive"],
-    "পেট": ["gastroenterologist", "stomach", "digestive"],
+    "gastro": ["gastrologist", "gastroenterologist", "gastroenterology", "gastro specialist", "stomach", "digestive"],
+    "gastroenterology": ["gastrologist", "gastroenterologist", "gastroenterology", "gastro specialist", "stomach", "digestive"],
+    "gastrologist": ["gastrologist", "gastroenterologist", "gastroenterology", "gastro specialist", "stomach", "digestive"],
+    "গ্যাস্ট্রো": ["gastrologist", "gastroenterologist", "gastroenterology", "gastro specialist", "stomach", "digestive"],
+    "পেট": ["gastrologist", "gastroenterologist", "gastroenterology", "gastro specialist", "stomach", "digestive"],
     # DB category uses "Gastrologist" in some datasets; include common spellings.
     "গ্যাস্ট্রোএন্টারোলজি": [
         "gastrologist",
         "gastroenterologist",
         "gastroenterology",
         "gastrology",
-        "gastro",
+    "gastro",
     ],
 }
 
+SPECIALTY_INTENT_WORDS = {
+    "cardiologist", "cardiology", "neurologist", "neurology", "orthopedics",
+    "orthopedic", "orthopaedic", "orthopaedics", "gastrologist",
+    "gastroenterologist", "gastroenterology", "dentist", "eye specialist",
+    "eye doctor", "ophthalmologist", "ophthalmology", "child specialist",
+    "pediatrician", "pediatrics", "ent", "nephrologist", "pulmonologist",
+    "dermatologist", "gynecologist", "gynaecologist",
+    "কার্ডিও", "হৃদরোগ", "নিউরো", "স্নায়ু", "অর্থো", "হাড়", "গ্যাস্ট্রো",
+    "চক্ষু", "চোখ", "আই", "শিশু", "পেডিয়াট্রিক", "দাঁত", "ডেন্টাল",
+}
+
 
 def _normalize_day(term: str) -> str:
     raw = _clean_text(term)
@@ -307,6 +338,74 @@ def _expand_search_terms(text: str) -> list[str]:
     return sorted(terms)
 
 
+def _normalize_lookup_text(text: str) -> str:
+    raw = _clean_text(text).lower()
+    if not raw:
+        return ""
+    raw = raw.replace("ডাঃ", "").replace("ডা.", "").replace("dr.", "").replace("dr", "")
+    raw = re.sub(r"[^0-9a-z\u0980-\u09ff]+", "", raw)
+    return raw
+
+
+def _doctor_search_score(row: dict, terms: list[str], day_text: str = "") -> float:
+    name = _clean_text(row.get("doctor_name", "")).lower()
+    category = _clean_text(row.get("category", "")).lower()
+    days = _clean_text(row.get("visiting_days", "")).lower()
+    haystacks = [
+        name,
+        category,
+        days,
+        _normalize_lookup_text(name),
+        _normalize_lookup_text(category),
+        _normalize_lookup_text(days),
+    ]
+
+    score = 0.0
+    if day_text and day_text.lower() in days:
+        score += 3.0
+
+    for term in terms:
+        norm_term = _normalize_lookup_text(term)
+        if not norm_term:
+            continue
+        if any(norm_term in hay for hay in haystacks if hay):
+            score += 3.0
+            continue
+        best = 0.0
+        for hay in haystacks:
+            if not hay:
+                continue
+            best = max(best, SequenceMatcher(None, norm_term, hay).ratio())
+        if best >= 0.72:
+            score += 1.5
+        elif best >= 0.62:
+            score += 0.5
+
+    return score
+
+
+async def _fallback_doctor_search(
+    db_path: str,
+    terms: list[str],
+    day_text: str = "",
+    limit: int = 10,
+) -> list[dict]:
+    async with aiosqlite.connect(db_path) as db:
+        db.row_factory = aiosqlite.Row
+        cursor = await db.execute("SELECT * FROM doctors")
+        rows = await cursor.fetchall()
+
+    scored: list[tuple[float, dict]] = []
+    for row in rows:
+        row_dict = dict(row)
+        score = _doctor_search_score(row_dict, terms, day_text=day_text)
+        if score > 0:
+            scored.append((score, row_dict))
+
+    scored.sort(key=lambda item: (-item[0], _clean_text(item[1].get("doctor_name", ""))))
+    return [row for _, row in scored[:limit]]
+
+
 def _parse_visit_date(text: str) -> Optional[str]:
     """
     Parse a user-facing date into YYYY-MM-DD in Bangladesh time.
@@ -417,6 +516,8 @@ def _looks_like_tool_turn(text: str) -> bool:
     if not lowered:
         return False
     return any(token.lower() in lowered for token in TOOL_INTENT_WORDS) or any(
+        token.lower() in lowered for token in SPECIALTY_INTENT_WORDS
+    ) or any(
         token.lower() in lowered for token in BOOKING_CONFIRM_WORDS
     )
 
@@ -690,6 +791,9 @@ async def find_doctors(query: str = "", visiting_day: str = "") -> str:
         cursor = await db.execute(sql, params)
         rows = await cursor.fetchall()
 
+    if not rows:
+        rows = await _fallback_doctor_search(db_path, terms, day_text=day_text)
+
     if not rows:
         return json.dumps({
             "success": False,
@@ -754,6 +858,10 @@ async def search_doctor(
         cursor = await db.execute(query, params)
         rows   = await cursor.fetchall()
 
+    if not rows:
+        fallback_terms = sorted(set(name_terms + category_terms + ([day_text] if day_text else [])))
+        rows = await _fallback_doctor_search(db_path, fallback_terms, day_text=day_text)
+
     if not rows:
         return json.dumps({"success": False, "message": "No doctors found.", "data": []}, ensure_ascii=False)
 
@@ -1345,6 +1453,10 @@ UPDATE / CANCEL FLOW (important):
 TOOL RULES:
 - Use `find_doctors` first for doctor search, specialty search, and availability search.
 - Use `get_doctors_by_day` or `get_categories_by_day` when the user asks about a day directly.
+- If the user only says a specialty, doctor type, or availability phrase like
+  "Neurologist", "cardiologist", "eye specialist", "child specialist",
+  "orthopedics", "নিউরোলজি", "চক্ষু", or "শিশু", treat it as a doctor search
+  request and call a tool instead of answering from memory.
 - Use `book_appointment` only after identifying the doctor and required patient details.
 - Use `update_appointment` when the user wants to change an existing appointment.
 - Never invent `doctor_id`. Get it from tool results or resolve by doctor_name/category.
@@ -1361,12 +1473,24 @@ LANGUAGE RULE
 - Number & Format Rules:
     - Show numbers in Bangla digits (০-৯) when responding in Bangla.
     - Avoid mixing English digits in Bangla sentences unless required technically.
-- Time & Date Format (spoken Bangla style):
+- Time Format (spoken Bangla style):
     - Use natural spoken expressions:
         - "দশটা ২৮ মিনিট"
         - "চারটা বেজে তিরিশ মিনিট"
         - "এখন টাইম হচ্ছে সাতটা তিরিশ"
 
+
+- Date Format (spoken Bangla style):
+    - Use natural spoken expressions:
+        - "আজকে বারোই নভেম্বর।"
+        - "আজকে পাঁচই মার্চ।"
+        - "আজকে বাইশেই জুন।"
+        - "জানুয়ারি মাসের আঠারো তারিখ"
+        - "ফেব্রুয়ারি মাসের ছয় তারিখ"
+        - "সেপ্টেম্বরের চার তারিখ"
+        - "মে মাসের বিশ তারিখ"
+
+
 - Year Format (spoken Bangla style):
     - "দুই হাজার পঁচিশ সাল"
     - "উনিশশো একাত্তর সাল"
@@ -1419,6 +1543,8 @@ The previous assistant turn failed to use a tool even though the user intent is
 You must now choose the correct tool instead of answering in prose:
 - Use `find_doctors` or `search_doctor` for doctor/specialty/availability questions.
 - Use `get_doctors_by_day` or `get_categories_by_day` for day-based availability.
+- If the user says only a specialty or doctor type, or asks which doctors are
+  available, call a search tool immediately. Do not answer from memory.
 - Use `book_appointment` when the user is confirming a booking.
 - Use `update_appointment` when the user wants to update an appointment.
 - Use `delete_appointment` when the user is cancelling a booking.

frontend/script.js

@@ -107,7 +107,7 @@ let _currentTurn = 0;
 // Client-side playback speed multiplier.
 // This makes speech faster immediately even if the TTS provider speed setting
 // is limited/ignored. 1.0 = normal, >1.0 = faster.
-let TTS_PLAYBACK_RATE = 1.025;
+let TTS_PLAYBACK_RATE = 1.0;
 let brainMode = false;
 let brainVoiceActive = false;
 let brainRestartTimer = null;
@@ -117,64 +117,41 @@ let voicePendingPackets = [];
 let brainLastResponse = '';
 let _brainWelcomed = false;
 
-const SPOKEN_DIGIT_WORDS = {
-  '0': 'শূন্য',
-  '1': 'এক',
-  '2': 'দুই',
-  '3': 'তিন',
-  '4': 'চার',
-  '5': 'পাঁচ',
-  '6': 'ছয়',
-  '7': 'সাত',
-  '8': 'আট',
-  '9': 'নয়',
-  '০': 'শূন্য',
-  '১': 'এক',
-  '২': 'দুই',
-  '৩': 'তিন',
-  '৪': 'চার',
-  '৫': 'পাঁচ',
-  '৬': 'ছয়',
-  '৭': 'সাত',
-  '৮': 'আট',
-  '৯': 'নয়',
-  '٠': 'শূন্য',
-  '١': 'এক',
-  '٢': 'দুই',
-  '٣': 'তিন',
-  '٤': 'চার',
-  '٥': 'পাঁচ',
-  '٦': 'ছয়',
-  '٧': 'সাত',
-  '٨': 'আট',
-  '٩': 'নয়',
+const VISIBLE_DIGIT_MAP = {
+  '০': '0',
+  '১': '1',
+  '২': '2',
+  '৩': '3',
+  '৪': '4',
+  '৫': '5',
+  '৬': '6',
+  '৭': '7',
+  '৮': '8',
+  '৯': '9',
+  '٠': '0',
+  '١': '1',
+  '٢': '2',
+  '٣': '3',
+  '٤': '4',
+  '٥': '5',
+  '٦': '6',
+  '٧': '7',
+  '٨': '8',
+  '٩': '9',
 };
 
-function _spokenDigitWords(chunk) {
-  const digits = Array.from(chunk).filter((ch) => ch in SPOKEN_DIGIT_WORDS);
-  if (digits.length < 10) return chunk;
-  return digits.map((ch) => SPOKEN_DIGIT_WORDS[ch]).join(' ');
+function _toAsciiDigits(text) {
+  return String(text || '').replace(
+    /[০-৯٠-٩]/g,
+    (ch) => VISIBLE_DIGIT_MAP[ch] || ch,
+  );
 }
 
 function _normalizeVisibleAiText(text) {
   if (!text) return '';
-  let out = String(text)
-    .replaceAll('উপলব্ধ', 'এভেলেবেল')
-    .replaceAll('জ্বি', 'আচ্ছা');
-  out = out.replace(
-    /[+\d০-৯٠-٩][\d০-৯٠-٩\s().\-]{8,}[\d০-৯٠-٩]/g,
-    (match, offset, whole) => {
-      const spoken = _spokenDigitWords(match);
-      if (spoken === match) return match;
-      const prev = offset > 0 ? whole[offset - 1] : '';
-      const next = offset + match.length < whole.length ? whole[offset + match.length] : '';
-      let value = spoken;
-      if (prev && !/\s/.test(prev) && !/[([<{\"']/.test(prev)) value = ' ' + value;
-      if (next && !/\s/.test(next) && !/[\])>.,!?;:}\"']/.test(next)) value = value + ' ';
-      return value;
-    },
+  return _toAsciiDigits(
+    String(text).replaceAll('উপলব্ধ', 'এভেলেবেল').replaceAll('জ্বি', 'আচ্ছা'),
   );
-  return out.replace(/[ \t]{2,}/g, ' ');
 }
 
 const BRAIN_WELCOME_TEXT =
@@ -376,12 +353,10 @@ function onVoiceMsg(ev) {
     // We buffer/reorder by seq inside a turn, and ignore late packets from older turns.
     const u8 = new Uint8Array(ev.data);
     if (u8.length <= 8) return;
-    const turn =
-      (u8[0] << 24) | (u8[1] << 16) | (u8[2] << 8) | (u8[3] << 0);
-    const seq =
-      (u8[4] << 24) | (u8[5] << 16) | (u8[6] << 8) | (u8[7] << 0);
+    const turn = (u8[0] << 24) | (u8[1] << 16) | (u8[2] << 8) | (u8[3] << 0);
+    const seq = (u8[4] << 24) | (u8[5] << 16) | (u8[6] << 8) | (u8[7] << 0);
     const turnU = turn >>> 0;
-    if (turnU !== (_currentTurn >>> 0)) return;
+    if (turnU !== _currentTurn >>> 0) return;
     const payload = ev.data.slice(8);
     _pendingAudio.set(seq >>> 0, payload);
 
@@ -553,7 +528,7 @@ function _renderAiText(force = false) {
 }
 
 function _setCaption(text) {
-  _captionText = text || '';
+  _captionText = _normalizeVisibleAiText(text);
   if (_captionRaf) return;
   _captionRaf = requestAnimationFrame(() => {
     _captionRaf = 0;
@@ -1229,10 +1204,11 @@ function appendMsg(text, who) {
   if (brainMode && who === 'user') return null;
   const d = document.createElement('div');
   d.className = 'message ' + who;
+  const visibleText = _normalizeVisibleAiText(text);
   if (who === 'ai' && typeof marked !== 'undefined') {
-    d.innerHTML = marked.parse(text || '');
+    d.innerHTML = marked.parse(visibleText || '');
   } else {
-    d.textContent = text;
+    d.textContent = visibleText;
   }
   chatBox.appendChild(d);
   chatBox.scrollTop = chatBox.scrollHeight;
@@ -1266,7 +1242,8 @@ function setBrainMode(on) {
   if (voiceCaption) voiceCaption.textContent = '';
   if (brainMode) {
     brainBubbleSttText.textContent = 'Listening…';
-    brainBubbleTtsText.textContent = brainLastResponse || 'Waiting…';
+    brainBubbleTtsText.textContent =
+      _normalizeVisibleAiText(brainLastResponse) || 'Waiting…';
     brainVoiceActive = true;
     sidebarEl.classList.add('collapsed');
     sidebarToggle.textContent = '›';
@@ -1336,14 +1313,14 @@ function _brainModeSetSearch(active) {
 
 function _brainSetSttBubble(text) {
   if (!brainBubbleStt || !brainBubbleSttText) return;
-  const value = (text || '').trim();
+  const value = _normalizeVisibleAiText(text).trim();
   brainBubbleSttText.textContent = value || 'Listening…';
   brainBubbleStt.classList.toggle('active', !!value);
 }
 
 function _brainSetTtsBubble(text, active = true) {
   if (!brainBubbleTts || !brainBubbleTtsText) return;
-  const value = (text || '').trim();
+  const value = _normalizeVisibleAiText(text).trim();
   brainBubbleTtsText.textContent = value || 'Waiting…';
   brainBubbleTts.classList.toggle('active', !!value || !!active);
   brainBubbleTts.classList.toggle('speaking', !!active);