shaktideadmin commited on
Commit
bb4a2ce
Β·
verified Β·
1 Parent(s): 029bd6c

Update main.py

Browse files
Files changed (1) hide show
  1. main.py +465 -36
main.py CHANGED
@@ -434,17 +434,30 @@ MEDICAL_KB = {
434
  }
435
 
436
  # ── #20: A/B Testing Variants ────────────────────────────────────────────────
437
- EXTRACT_SYSTEM_V2 = (
438
- 'You are a clinical symptom extractor for a women\'s menstrual health app.\n'
439
- 'The patient may write in Hindi, English, or Hinglish.\n'
440
  'Rule-based extraction already found some symptoms.\n\n'
441
- 'Your job:\n'
442
- '1. Add SYMPTOMS ONLY that are EXPLICITLY mentioned or very strongly implied. NEVER infer. NEVER add disease names.\n'
443
- '2. Also extract SEVERITY for each: "mild", "moderate", "severe", or "unknown".\n'
444
- '3. Ask UP TO 2 questions. If reproductive symptoms exist, ONE must be about cycle details. Combine cycle length and last period into one question. Add "(you can say I don\'t know or skip)".\n'
445
- '4. Include an "acknowledgment" field β€” a brief warm sentence acknowledging the symptoms the patient described.\n'
446
- 'Respond ONLY with compact JSON:\n'
447
- '{"additional_symptoms":[{"name":"pelvic_pain","severity":"moderate"}],"acknowledgment":"I understand you\'re experiencing pelvic pain.","reproductive_context":"menstrual","questions":[{"id":"q1","text":"How long is your usual cycle and when did your last period start? (you can say I don\'t know or skip)"}]}'
 
 
 
 
 
 
 
 
 
 
 
 
 
448
  )
449
 
450
  # ══════════════════════════════════════════════════════════════════════════════
@@ -462,7 +475,8 @@ PHRASE_MAP = {
462
  r"loss of sensation|numbness|numb": "loss_of_sensation",
463
  r"visual aura|aura|spots.*vision|vision disturbance|blurred vision|dhundla": "spots_or_clouds_in_vision",
464
  r"sensitivity.*light|light.*sensitiv|photophobia|light.*hurts|bright.*light|sensitive.*light": "photophobia",
465
- r"tummy.*killing|killing.*tummy|stomach.*killing|tummy.*hurt|stomach.*hurt|stomachache|tummy.*ache|tummy.*pain|stomach.*ache|belly.*pain|belly.*hurt|abdominal pain|lower belly|lower tummy|pet.*dard|pet mein|pait|paet": "lower_abdominal_pain",
 
466
  r"upper.*stomach|upper.*abdom|upper.*belly": "upper_abdominal_pain",
467
  r"sharp.*stomach|sharp.*abdom|sharp.*belly": "sharp_abdominal_pain",
468
  r"burning.*stomach|burning.*abdom|acid.*stomach|jalan.*pet": "burning_abdominal_pain",
@@ -536,6 +550,90 @@ PHRASE_MAP = {
536
  r"hair loss|hair fall|losing hair|hair.*falling|baal jhad|baal girna|baal toot": "irregular_appearing_scalp",
537
  }
538
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
539
  # ══════════════════════════════════════════════════════════════════════════════
540
  # 7. PYDANTIC MODELS
541
  # ══════════════════════════════════════════════════════════════════════════════
@@ -745,31 +843,6 @@ async def _groq_text(system: str, user: str, max_tokens: int = 500, temperature:
745
  # ══════════════════════════════════════════════════════════════════════════════
746
  # 10. SYSTEM PROMPTS
747
  # ══════════════════════════════════════════════════════════════════════════════
748
- EXTRACT_SYSTEM = (
749
- 'You are a clinical symptom extractor for a women\'s menstrual health app.\n'
750
- 'The patient may write in Hindi, English, or Hinglish.\n'
751
- 'Rule-based extraction already found some symptoms.\n\n'
752
- 'Your job:\n'
753
- '1. Add SYMPTOMS ONLY that are EXPLICITLY mentioned or very strongly implied by the patient.\n'
754
- ' NEVER infer symptoms the patient did not describe β€” do not add "likely" symptoms.\n'
755
- ' - ONLY ask about cycle/periods if symptoms EXPLICITLY include: unpredictable_menstruation, painful_menstruation, heavy_menstrual_flow, pelvic_pain, vaginal_discharge, infertility, or intermenstrual_bleeding. NEVER ask cycle questions for headache, fever, nausea, vomiting, or stomach pain.\n'
756
- ' NEVER add disease names (not "PCOS", "endometriosis", "migraine" etc.)\n'
757
- '2. Ask UP TO 2 questions. IMPORTANT RULES for questions:\n'
758
- 'CRITICAL: When asking about a symptom the user already mentioned, you MUST reference it as "the [symptom] you mentioned" or "this [symptom]". NEVER ask "Do you have [symptom]?" if they just told you they have it.',
759
- ' - ONLY ask about cycle/periods if symptoms EXPLICITLY include: unpredictable_menstruation, painful_menstruation, heavy_menstrual_flow, pelvic_pain, vaginal_discharge, infertility, intermenstrual_bleeding, or hirsutism. NEVER ask cycle questions for general symptoms like headache, fever, nausea, vomiting, or stomach pain.\n'
760
- ' - For purely general symptoms with NO menstrual symptoms, do NOT ask cycle questions and set reproductive_context to null.\n'
761
- ' - Second question (optional) covers the most likely missing symptom cluster.\n'
762
- ' - Always add "(you can say I don\'t know or skip)" to each question.\n'
763
- ' - NEVER ask about both cycle length and last period in separate questions β€” combine them.\n'
764
- ' - If cycle_data_already_provided is true, do NOT ask any cycle questions.\n'
765
- ' - Do NOT ask about symptoms already in the found list.\n'
766
- ' - Always ask questions in English only.\n'
767
- ' - Include an "acknowledgment" field β€” a brief, warm sentence that first acknowledges the symptoms the patient has already mentioned. Example: "I understand you\'re experiencing pelvic pain and hair thinning." This makes the conversation feel natural and heard.\n\n'
768
- 'Respond ONLY with compact JSON on one line, no markdown:\n'
769
- '{"additional_symptoms":["pelvic_pain","hair_thinning"],"acknowledgment":"I understand you\'re experiencing pelvic pain and hair thinning.","reproductive_context":"menstrual",'
770
- '"questions":[{"id":"q1","text":"How long is your usual cycle and when did your last period start? (you can say I don\'t know or skip)"}]}'
771
- )
772
-
773
  SYNTHESIZE_SYSTEM = (
774
  'You are a clinical routing engine for a women\'s menstrual health app.\n'
775
  'Given patient message, extracted symptoms, and follow-up answers:\n\n'
@@ -862,6 +935,229 @@ RAG_SYSTEM = (
862
  'Rules: Plain language, 3-4 sentences max. No markdown.'
863
  )
864
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
865
  # ══════════════════════════════════════════════════════════════════════════════
866
  # 11. LLM CALLS
867
  # ══════════════════════════════════════════════════════════════════════════════
@@ -1100,6 +1396,139 @@ def _check_daily_limit(identifier: str) -> bool:
1100
  _daily_usage[key] = count + 1
1101
  return True
1102
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1103
  # ══════════════════════════════════════════════════════════════════════════════
1104
  # 14. MAIN CHAT ENDPOINT
1105
  # ══════════════════════════════════════════════════════════════════════════════
 
434
  }
435
 
436
  # ── #20: A/B Testing Variants ────────────────────────────────────────────────
437
+ EXTRACT_SYSTEM = (
438
+ 'You are a clinical symptom extractor for a health app.\n'
439
+ 'The patient writes in English.\n'
440
  'Rule-based extraction already found some symptoms.\n\n'
441
+
442
+ '═══ STRICT RULES ═══\n'
443
+ '1. You MUST respond in English only. No Hindi, no Hinglish.\n'
444
+ '2. NEVER ask about any symptom already in found_symptoms.\n'
445
+ '3. Every question MUST start with "Do you also have" or "Are you also experiencing".\n'
446
+ '4. NEVER ask about periods/cycles unless found_symptoms contains: unpredictable_menstruation, painful_menstruation, heavy_menstrual_flow, pelvic_pain, vaginal_discharge, infertility, intermenstrual_bleeding, hirsutism.\n'
447
+ '5. For stomach pain, nausea, fever, headache, cough, fatigue β†’ do NOT ask about periods.\n'
448
+ '6. NEVER include disease names in additional_symptoms.\n'
449
+ '7. NEVER infer symptoms not explicitly mentioned.\n\n'
450
+
451
+ '═══ OUTPUT FORMAT ═══\n'
452
+ 'Return ONLY this JSON on one line:\n'
453
+ '{"acknowledgment":"I understand you are experiencing [list all symptoms from found_symptoms in plain English].","additional_symptoms":[],"reproductive_context":null,"questions":[{"id":"q1","text":"Do you also have [new symptom]? (you can say I dont know or skip)"}]}\n\n'
454
+
455
+ '═══ EXAMPLES ═══\n'
456
+ 'Input: found_symptoms=["upper_abdominal_pain","nausea","fever"]\n'
457
+ 'Output: {"acknowledgment":"I understand you are experiencing stomach pain, nausea, and fever.","additional_symptoms":[],"reproductive_context":null,"questions":[{"id":"q1","text":"Do you also have vomiting or loose motions? (you can say I dont know or skip)"}]}\n\n'
458
+
459
+ 'Input: found_symptoms=["unpredictable_menstruation","acne_or_pimples","weight_gain"]\n'
460
+ 'Output: {"acknowledgment":"I understand you are experiencing irregular periods, acne, and weight gain.","additional_symptoms":[],"reproductive_context":"menstrual","questions":[{"id":"q1","text":"Do you also have facial hair growth or hair thinning? (you can say I dont know or skip)"},{"id":"q2","text":"What is your usual cycle length and when did your last period start? (you can say I dont know or skip)"}]}'
461
  )
462
 
463
  # ══════════════════════════════════════════════════════════════════════════════
 
475
  r"loss of sensation|numbness|numb": "loss_of_sensation",
476
  r"visual aura|aura|spots.*vision|vision disturbance|blurred vision|dhundla": "spots_or_clouds_in_vision",
477
  r"sensitivity.*light|light.*sensitiv|photophobia|light.*hurts|bright.*light|sensitive.*light": "photophobia",
478
+ r"stomach.*killing|killing.*stomach|stomach.*hurt|stomachache|stomach.*pain|stomach.*ache|tummy.*killing|killing.*tummy|tummy.*hurt|tummy.*ache|tummy.*pain|pait|paet|pet.*dard|pet mein|belly.*pain|belly.*hurt|abdominal pain": "upper_abdominal_pain",
479
+ r"lower belly|lower tummy|lower abdomen": "lower_abdominal_pain",
480
  r"upper.*stomach|upper.*abdom|upper.*belly": "upper_abdominal_pain",
481
  r"sharp.*stomach|sharp.*abdom|sharp.*belly": "sharp_abdominal_pain",
482
  r"burning.*stomach|burning.*abdom|acid.*stomach|jalan.*pet": "burning_abdominal_pain",
 
550
  r"hair loss|hair fall|losing hair|hair.*falling|baal jhad|baal girna|baal toot": "irregular_appearing_scalp",
551
  }
552
 
553
+ SYMPTOM_TO_ENGLISH = {
554
+ "headache": "headache",
555
+ "frontal_headache": "headache in the front of your head",
556
+ "dizziness": "dizziness",
557
+ "fainting": "fainting",
558
+ "seizures": "seizures",
559
+ "disturbance_of_memory": "memory issues",
560
+ "difficulty_speaking": "difficulty speaking",
561
+ "ringing_in_ear": "ringing in your ears",
562
+ "loss_of_sensation": "numbness or loss of sensation",
563
+ "spots_or_clouds_in_vision": "spots or blurriness in your vision",
564
+ "photophobia": "sensitivity to light",
565
+ "upper_abdominal_pain": "stomach pain",
566
+ "lower_abdominal_pain": "lower abdominal pain",
567
+ "sharp_abdominal_pain": "sharp abdominal pain",
568
+ "burning_abdominal_pain": "burning sensation in your stomach",
569
+ "nausea": "nausea",
570
+ "vomiting": "vomiting",
571
+ "diarrhea": "diarrhea or loose motions",
572
+ "constipation": "constipation",
573
+ "stomach_bloating": "bloating",
574
+ "heartburn": "heartburn",
575
+ "decreased_appetite": "loss of appetite",
576
+ "fever": "fever",
577
+ "chills": "chills",
578
+ "fatigue": "fatigue",
579
+ "weakness": "weakness",
580
+ "ache_all_over": "body ache",
581
+ "sweating": "sweating",
582
+ "feeling_ill": "feeling unwell",
583
+ "flu-like_syndrome": "flu-like symptoms",
584
+ "cough": "cough",
585
+ "shortness_of_breath": "shortness of breath",
586
+ "sore_throat": "sore throat",
587
+ "nasal_congestion": "blocked or runny nose",
588
+ "wheezing": "wheezing",
589
+ "chest_tightness": "tightness in your chest",
590
+ "sharp_chest_pain": "chest pain",
591
+ "burning_chest_pain": "burning sensation in your chest",
592
+ "blood_clots_during_menstrual_periods": "blood clots during your periods",
593
+ "heavy_menstrual_flow": "heavy menstrual bleeding",
594
+ "unpredictable_menstruation": "irregular periods",
595
+ "painful_menstruation": "painful periods",
596
+ "vaginal_discharge": "vaginal discharge",
597
+ "hot_flashes": "hot flashes",
598
+ "pelvic_pain": "pelvic pain",
599
+ "infertility": "difficulty getting pregnant",
600
+ "long_menstrual_periods": "periods that last longer than usual",
601
+ "frequent_menstruation": "frequent periods",
602
+ "intermenstrual_bleeding": "bleeding between periods",
603
+ "hirsutism": "facial hair growth",
604
+ "hair_thinning": "hair thinning",
605
+ "painful_intercourse": "pain during intercourse",
606
+ "oily_skin": "oily skin",
607
+ "weight_loss": "weight loss",
608
+ "skin_rash": "skin rash",
609
+ "itching_of_skin": "itching",
610
+ "jaundice": "jaundice",
611
+ "acne_or_pimples": "acne",
612
+ "skin_dryness__peeling__scaliness__or_roughness": "dry skin",
613
+ "joint_pain": "joint pain",
614
+ "back_pain": "back pain",
615
+ "low_back_pain": "lower back pain",
616
+ "neck_pain": "neck pain",
617
+ "knee_pain": "knee pain",
618
+ "shoulder_pain": "shoulder pain",
619
+ "palpitations": "palpitations",
620
+ "irregular_heartbeat": "irregular heartbeat",
621
+ "increased_heart_rate": "increased heart rate",
622
+ "frequent_urination": "frequent urination",
623
+ "painful_urination": "burning during urination",
624
+ "blood_in_urine": "blood in urine",
625
+ "pain_in_eye": "eye pain",
626
+ "eye_redness": "redness in your eyes",
627
+ "diminished_vision": "blurred vision",
628
+ "anxiety_and_nervousness": "anxiety",
629
+ "depression": "depression",
630
+ "insomnia": "difficulty sleeping",
631
+ "restlessness": "restlessness",
632
+ "excessive_anger": "irritability",
633
+ "weight_gain": "weight gain",
634
+ "irregular_appearing_scalp": "hair loss",
635
+ }
636
+
637
  # ══════════════════════════════════════════════════════════════════════════════
638
  # 7. PYDANTIC MODELS
639
  # ══════════════════════════════════════════════════════════════════════════════
 
843
  # ══════════════════════════════════════════════════════════════════════════════
844
  # 10. SYSTEM PROMPTS
845
  # ══════════════════════════════════════════════════════════════════════════════
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
846
  SYNTHESIZE_SYSTEM = (
847
  'You are a clinical routing engine for a women\'s menstrual health app.\n'
848
  'Given patient message, extracted symptoms, and follow-up answers:\n\n'
 
935
  'Rules: Plain language, 3-4 sentences max. No markdown.'
936
  )
937
 
938
+
939
+ # ══════════════════════════════════════════════════════════════════════════════
940
+ # 11. CHAT ENDPOINT
941
+ # ══════════════════════════════════════════════════════════════════════════════
942
+
943
+ class ChatRequest(BaseModel):
944
+ conversation_id: Optional[str] = None
945
+ accumulated_symptoms: List[str] = []
946
+ accumulated_negated: List[str] = []
947
+ asked_questions: List[str] = []
948
+ cycle_data: Dict[str, Any] = {}
949
+ pending_questions: List[dict] = []
950
+ conversation_turn: int = 0
951
+ chat_history: List[dict] = []
952
+ first_message: Optional[str] = None
953
+
954
+ @app.post("/api/chat")
955
+ async def chat_endpoint(req: ChatRequest, request: Request):
956
+ user_message = req.first_message if req.conversation_turn == 0 else req.chat_history[-1]["content"] if req.chat_history else ""
957
+
958
+ if not user_message or not user_message.strip():
959
+ return {"error": "Empty message", "message": "Please describe your symptoms."}
960
+
961
+ # ── Step 1: AUP Check ──
962
+ for pattern in AUP_PATTERNS:
963
+ if re.search(pattern, user_message.lower()):
964
+ return {"error": "policy_violation", "message": "I cannot help with that request."}
965
+
966
+ # ── Step 2: Crisis Check ──
967
+ for kw in CRISIS_KEYWORDS:
968
+ if kw in user_message.lower():
969
+ return {
970
+ "acknowledgment": "🚨 This sounds like a medical emergency. Please go to the nearest hospital emergency room immediately.",
971
+ "questions": [],
972
+ "accumulated_symptoms": req.accumulated_symptoms,
973
+ "conversation_turn": req.conversation_turn,
974
+ "is_emergency": True
975
+ }
976
+
977
+ # ── Step 3: Red Flag Triage ──
978
+ triage = triage_check(user_message)
979
+ if triage:
980
+ level, response = triage
981
+ return {
982
+ "acknowledgment": response,
983
+ "questions": [],
984
+ "accumulated_symptoms": req.accumulated_symptoms,
985
+ "conversation_turn": req.conversation_turn,
986
+ "triage_level": level
987
+ }
988
+
989
+ # ── Step 4: Intent Classification ──
990
+ intent = await classify_intent(user_message)
991
+
992
+ # Handle greetings
993
+ if intent == "greeting":
994
+ return {
995
+ "acknowledgment": "Hello! I am ShakTide AI. Please describe your symptoms and I will help you understand what might be going on.",
996
+ "questions": [],
997
+ "accumulated_symptoms": req.accumulated_symptoms,
998
+ "conversation_turn": req.conversation_turn
999
+ }
1000
+
1001
+ # Handle gratitude
1002
+ if intent == "gratitude":
1003
+ return {
1004
+ "acknowledgment": "You are welcome! Is there anything else I can help you with?",
1005
+ "questions": [],
1006
+ "accumulated_symptoms": req.accumulated_symptoms,
1007
+ "conversation_turn": req.conversation_turn
1008
+ }
1009
+
1010
+ # Handle irrelevant
1011
+ if intent == "irrelevant":
1012
+ return {
1013
+ "acknowledgment": "I am here to help with health concerns. Please describe any symptoms you are experiencing.",
1014
+ "questions": [],
1015
+ "accumulated_symptoms": req.accumulated_symptoms,
1016
+ "conversation_turn": req.conversation_turn
1017
+ }
1018
+
1019
+ # ── Step 5: Extract Symptoms (Rule-based) ──
1020
+ rule_based_symptoms = rule_extract_symptoms(user_message)
1021
+
1022
+ # ── Step 6: Handle Follow-up Answers ──
1023
+ if intent == "followup_answer" and req.pending_questions:
1024
+ # Extract symptoms from the user's answer
1025
+ answer_symptoms = rule_extract_symptoms(user_message)
1026
+
1027
+ # Check for "no" / "none" / negation
1028
+ answer_lower = user_message.lower().strip()
1029
+ negated_this_turn = []
1030
+
1031
+ if answer_lower in {"no", "none", "nope", "not really", "nothing else", "no other symptoms", "thats all", "that's all", "only that"}:
1032
+ # User said no more symptoms - proceed to diagnosis
1033
+ all_symptoms = list(set(req.accumulated_symptoms))
1034
+
1035
+ if len(all_symptoms) >= MIN_SYMPTOMS:
1036
+ diagnosis = await run_diagnosis(all_symptoms, req.cycle_data)
1037
+ return {
1038
+ "acknowledgment": "Thank you for that information. Based on your symptoms, here is my analysis.",
1039
+ "questions": [],
1040
+ "accumulated_symptoms": all_symptoms,
1041
+ "conversation_turn": req.conversation_turn + 1,
1042
+ "diagnosis": diagnosis
1043
+ }
1044
+ else:
1045
+ return {
1046
+ "acknowledgment": "Thank you. I have limited information so far. Could you describe any other symptoms you are experiencing?",
1047
+ "questions": [],
1048
+ "accumulated_symptoms": all_symptoms,
1049
+ "conversation_turn": req.conversation_turn + 1
1050
+ }
1051
+
1052
+ # User answered with new symptoms
1053
+ all_symptoms = list(set(req.accumulated_symptoms + answer_symptoms))
1054
+
1055
+ # Build acknowledgment from new symptoms
1056
+ new_english = [SYMPTOM_TO_ENGLISH.get(s, s.replace("_", " ")) for s in answer_symptoms]
1057
+ if new_english:
1058
+ if len(new_english) == 1:
1059
+ ack = f"I understand you are also experiencing {new_english[0]}."
1060
+ elif len(new_english) == 2:
1061
+ ack = f"I understand you are also experiencing {new_english[0]} and {new_english[1]}."
1062
+ else:
1063
+ ack = f"I understand you are also experiencing {', '.join(new_english[:-1])}, and {new_english[-1]}."
1064
+ else:
1065
+ ack = "Thank you for that information."
1066
+
1067
+ # Check if we have enough symptoms for diagnosis
1068
+ if len(all_symptoms) >= MIN_SYMPTOMS:
1069
+ diagnosis = await run_diagnosis(all_symptoms, req.cycle_data)
1070
+ return {
1071
+ "acknowledgment": ack,
1072
+ "questions": [],
1073
+ "accumulated_symptoms": all_symptoms,
1074
+ "conversation_turn": req.conversation_turn + 1,
1075
+ "diagnosis": diagnosis
1076
+ }
1077
+ else:
1078
+ # Ask more questions
1079
+ followup_questions = await generate_followup_questions(all_symptoms, req.asked_questions, req.cycle_data)
1080
+ return {
1081
+ "acknowledgment": ack,
1082
+ "questions": followup_questions,
1083
+ "accumulated_symptoms": all_symptoms,
1084
+ "conversation_turn": req.conversation_turn + 1,
1085
+ "pending_questions": followup_questions
1086
+ }
1087
+
1088
+ # ── Step 7: First Symptom Report - Call LLM ──
1089
+ if rule_based_symptoms:
1090
+ # Convert to English for acknowledgment
1091
+ symptom_english = [SYMPTOM_TO_ENGLISH.get(s, s.replace("_", " ")) for s in rule_based_symptoms]
1092
+
1093
+ # Build prompt for LLM
1094
+ extract_msg = (
1095
+ f"Patient said: {user_message}\n"
1096
+ f"found_symptoms: {json.dumps(rule_based_symptoms)}\n"
1097
+ f"found_symptoms_in_english: {json.dumps(symptom_english)}\n"
1098
+ f"cycle_data_already_provided: {bool(req.cycle_data)}\n"
1099
+ f"already_asked: {json.dumps(req.asked_questions)}"
1100
+ )
1101
+
1102
+ try:
1103
+ llm_raw = await _groq(EXTRACT_SYSTEM, extract_msg, 400)
1104
+ llm_data = json.loads(llm_raw)
1105
+
1106
+ acknowledgment = llm_data.get("acknowledgment", "")
1107
+ additional_symptoms = llm_data.get("additional_symptoms", [])
1108
+ questions = llm_data.get("questions", [])
1109
+ repro_context = llm_data.get("reproductive_context")
1110
+ except Exception as e:
1111
+ logger.error(f"LLM extraction failed: {e}")
1112
+ # Fallback acknowledgment
1113
+ if len(symptom_english) == 1:
1114
+ acknowledgment = f"I understand you are experiencing {symptom_english[0]}."
1115
+ elif len(symptom_english) == 2:
1116
+ acknowledgment = f"I understand you are experiencing {symptom_english[0]} and {symptom_english[1]}."
1117
+ else:
1118
+ ack_list = ", ".join(symptom_english[:-1]) + ", and " + symptom_english[-1]
1119
+ acknowledgment = f"I understand you are experiencing {ack_list}."
1120
+
1121
+ additional_symptoms = []
1122
+ questions = []
1123
+ repro_context = None
1124
+
1125
+ # Combine symptoms
1126
+ all_symptoms = list(set(rule_based_symptoms + additional_symptoms))
1127
+
1128
+ # Filter out already asked questions
1129
+ asked_set = set(req.asked_questions)
1130
+ filtered_questions = [q for q in questions if q.get("text", "") not in asked_set]
1131
+
1132
+ # Update asked list
1133
+ new_asked = req.asked_questions + [q.get("text", "") for q in filtered_questions]
1134
+
1135
+ # Handle cycle data
1136
+ cycle_data = req.cycle_data.copy() if req.cycle_data else {}
1137
+ if repro_context == "menstrual" and not cycle_data:
1138
+ cycle_data = {}
1139
+
1140
+ return {
1141
+ "acknowledgment": acknowledgment,
1142
+ "questions": filtered_questions,
1143
+ "accumulated_symptoms": all_symptoms,
1144
+ "accumulated_negated": req.accumulated_negated,
1145
+ "conversation_id": req.conversation_id or str(datetime.datetime.now().timestamp()),
1146
+ "conversation_turn": req.conversation_turn + 1,
1147
+ "pending_questions": filtered_questions,
1148
+ "asked_questions": new_asked,
1149
+ "cycle_data": cycle_data,
1150
+ "message": acknowledgment # Fallback for old frontend
1151
+ }
1152
+
1153
+ # ── Step 8: No Symptoms Found ──
1154
+ return {
1155
+ "acknowledgment": "I want to help, but I did not catch any specific symptoms. Could you describe what you are feeling? For example: 'I have a headache and fever for 2 days.'",
1156
+ "questions": [],
1157
+ "accumulated_symptoms": req.accumulated_symptoms,
1158
+ "conversation_turn": req.conversation_turn + 1
1159
+ }
1160
+
1161
  # ══════════════════════════════════════════════════════════════════════════════
1162
  # 11. LLM CALLS
1163
  # ══════════════════════════════════════════════════════════════════════════════
 
1396
  _daily_usage[key] = count + 1
1397
  return True
1398
 
1399
+
1400
+ async def run_diagnosis(symptoms: list, cycle_data: dict = None) -> dict:
1401
+ """Run XGBoost prediction and return diagnosis card data."""
1402
+
1403
+ # Check medication overlap
1404
+ med_note = None # Would need medications from profile
1405
+
1406
+ # ── NLP Retrieval ──
1407
+ symptom_text = " ".join(symptoms).replace("_", " ")
1408
+
1409
+ tfidf_results = pd.DataFrame()
1410
+ semantic_results = pd.DataFrame()
1411
+
1412
+ if tfidf_retriever:
1413
+ try:
1414
+ tfidf_results = tfidf_retriever.query(symptom_text, top_k=5)
1415
+ except Exception as e:
1416
+ logger.error(f"TF-IDF failed: {e}")
1417
+
1418
+ if sbert_retriever:
1419
+ try:
1420
+ semantic_results = sbert_retriever.query(symptom_text, top_k=5)
1421
+ except Exception as e:
1422
+ logger.error(f"SBERT failed: {e}")
1423
+
1424
+ # Merge and rank candidates
1425
+ candidates = []
1426
+ seen = set()
1427
+ for df in [tfidf_results, semantic_results]:
1428
+ for _, row in df.iterrows():
1429
+ disease = row.get("disease", "")
1430
+ if disease and disease not in seen and not _is_blocked(disease, symptoms):
1431
+ seen.add(disease)
1432
+ candidates.append({"disease": disease, "score": float(row.get("score", 0))})
1433
+
1434
+ candidates.sort(key=lambda x: x["score"], reverse=True)
1435
+ candidate_names = [c["disease"] for c in candidates[:5]]
1436
+
1437
+ # ── XGBoost Prediction ──
1438
+ xgb_confidence = 0.0
1439
+ xgb_prediction = None
1440
+
1441
+ if disease_model and disease_encoder and symptom_scaler:
1442
+ try:
1443
+ # Create feature vector
1444
+ all_possible_symptoms = list(disease_encoder.classes_) if hasattr(disease_encoder, 'classes_') else []
1445
+ feature_vec = np.zeros(len(all_possible_symptoms))
1446
+ for s in symptoms:
1447
+ if s in all_possible_symptoms:
1448
+ idx = list(all_possible_symptoms).index(s)
1449
+ feature_vec[idx] = 1
1450
+
1451
+ feature_scaled = symptom_scaler.transform([feature_vec])
1452
+ probas = disease_model.predict_proba(feature_scaled)[0]
1453
+ max_idx = np.argmax(probas)
1454
+ xgb_prediction = disease_encoder.inverse_transform([max_idx])[0] if hasattr(disease_encoder, 'inverse_transform') else all_possible_symptoms[max_idx]
1455
+ xgb_confidence = float(probas[max_idx])
1456
+ except Exception as e:
1457
+ logger.error(f"XGBoost failed: {e}")
1458
+
1459
+ # ── LLM Verification ──
1460
+ final_condition = xgb_prediction or (candidate_names[0] if candidate_names else "Unknown condition")
1461
+ final_confidence = xgb_confidence
1462
+
1463
+ if candidate_names:
1464
+ verify_msg = (
1465
+ f"Confirmed symptoms: {json.dumps(symptoms)}\n"
1466
+ f"Candidates: {json.dumps(candidate_names)}\n"
1467
+ f"XGBoost predicted: {xgb_prediction} (confidence: {xgb_confidence})"
1468
+ )
1469
+ try:
1470
+ verify_raw = await _groq(VERIFY_SYSTEM, verify_msg, 200)
1471
+ verify_data = json.loads(verify_raw)
1472
+ final_condition = verify_data.get("final_condition", final_condition)
1473
+ conf_map = {"high": 0.85, "medium": 0.65, "low": 0.45}
1474
+ final_confidence = conf_map.get(verify_data.get("confidence", "medium"), xgb_confidence)
1475
+ except Exception as e:
1476
+ logger.error(f"Verify LLM failed: {e}")
1477
+
1478
+ # ── Check Exclusions ──
1479
+ exclusion_note = apply_exclusions(final_condition, symptoms)
1480
+
1481
+ # ── Build Symptom Lists ──
1482
+ matched = []
1483
+ unmatched = []
1484
+ disease_words = set(final_condition.lower().replace("_", " ").split())
1485
+ for s in symptoms:
1486
+ s_words = set(s.replace("_", " ").split())
1487
+ if disease_words & s_words or any(dw in s for dw in disease_words if len(dw) > 3):
1488
+ matched.append(s)
1489
+ else:
1490
+ unmatched.append(s)
1491
+
1492
+ # ── Severity ──
1493
+ if final_confidence >= 0.75:
1494
+ severity = "Moderate"
1495
+ sev_color = "#d97706"
1496
+ sev_dots = 3
1497
+ elif final_confidence >= 0.55:
1498
+ severity = "Low-Moderate"
1499
+ sev_color = "#23856b"
1500
+ sev_dots = 2
1501
+ else:
1502
+ severity = "Low"
1503
+ sev_color = "#52b788"
1504
+ sev_dots = 1
1505
+
1506
+ # ── Build Diet Recommendations ──
1507
+ eat_list = ["Light, easily digestible food", "Cooked vegetables and dal", "Fresh fruits like papaya or banana", "Small frequent meals", "Plenty of water and fluids"]
1508
+ avoid_list = ["Spicy and oily food", "Processed or packaged food", "Excessive tea or coffee", "Cold drinks and ice cream", "Heavy meals at night"]
1509
+
1510
+ # ── Build Precautions ──
1511
+ precautions = ["Rest adequately and get proper sleep", "Monitor your symptoms for any changes", "Stay hydrated with warm fluids", "Avoid self-medication", "Keep a symptom diary"]
1512
+
1513
+ return {
1514
+ "condition": final_condition,
1515
+ "summary": f"Based on your symptoms, this appears to be {final_condition.replace('_', ' ').title()}.",
1516
+ "why": f"Your symptoms of {', '.join([SYMPTOM_TO_ENGLISH.get(s, s.replace('_', ' ')) for s in matched[:3]])} align with this condition.",
1517
+ "severity": severity,
1518
+ "severity_color": sev_color,
1519
+ "severity_dots": sev_dots,
1520
+ "confidence": round(final_confidence * 100, 1),
1521
+ "matched_symptoms": [SYMPTOM_TO_ENGLISH.get(s, s.replace("_", " ")) for s in matched],
1522
+ "unmatched_symptoms": [SYMPTOM_TO_ENGLISH.get(s, s.replace("_", " ")) for s in unmatched],
1523
+ "eat": eat_list,
1524
+ "avoid": avoid_list,
1525
+ "precautions": precautions,
1526
+ "exclusion_note": exclusion_note,
1527
+ "candidates": candidate_names[:5],
1528
+ "ruled_out": [c for c in candidate_names[1:4] if c != final_condition]
1529
+ }
1530
+
1531
+
1532
  # ══════════════════════════════════════════════════════════════════════════════
1533
  # 14. MAIN CHAT ENDPOINT
1534
  # ══════════════════════════════════════════════════════════════════════════════