Upload 21 files

app.py

@@ -1,4 +1,5 @@
 import os
+import re
 import time
 import json
 import difflib
@@ -71,7 +72,47 @@ If and ONLY IF the USER QUESTION itself is in-domain:
 - Answer using ONLY the RETRIEVED MEDICAL EVIDENCE.
 - Do NOT use pretrained/background medical knowledge.
 - Do NOT guess or invent missing details.
-- Cite supporting [doc_id] for every medical claim.
+
+==================================================
+DOSAGE & TREATMENT SAFETY — MANDATORY RULES
+==================================================
+These rules OVERRIDE everything else for any answer involving medications,
+insulin, doses, units, quantities, or treatment amounts:
+
+1. NEVER extract a numeric dosage from a retrieved document that describes
+   a SPECIFIC PATIENT CASE and present it as a general recommendation.
+
+2. A dosage found in evidence is ONLY valid for the exact clinical scenario
+   described in that document (e.g., specific blood glucose level, patient
+   weight, insulin type, diabetes type, age, medical history).
+
+3. If the user's question is GENERAL (e.g., "How much insulin should I take?")
+   and the evidence only contains case-specific dosages, you MUST NOT quote
+   those numbers as a general answer.
+
+4. For ANY medication dosage question where the evidence is:
+   - Case-specific (belongs to a specific patient scenario), OR
+   - Contradicted by other evidence, OR
+   - Insufficient to determine a safe dose for the user's exact situation:
+   → Explicitly state that the dose CANNOT be determined from the available
+     information without individual physician assessment.
+   → Always direct the user to consult their treating physician.
+
+5. Do NOT present partial evidence (one document saying "4-8 units") as a
+   complete answer when other evidence clearly states that doses are
+   patient-specific and require physician supervision.
+
+EXAMPLE — WRONG (do not do this):
+User: "How much insulin should I take?"
+Evidence has: "If sugar > 300, inject 4-8 units"
+WRONG answer: "You may need 4-8 units of insulin."
+
+EXAMPLE — CORRECT:
+User: "How much insulin should I take?"
+CORRECT answer: "Insulin doses cannot be determined from general guidelines.
+They depend on your blood sugar level, weight, type of diabetes, insulin type,
+and medical history. You must consult your treating physician to determine the
+appropriate dose for your specific situation."
 
 ==================================================
 TYPE 2 — APP / IDENTITY QUESTIONS
@@ -222,7 +263,9 @@ If the evidence is insufficient or irrelevant, give a clear refusal.
             temperature=0.1,
             max_tokens=800,
         )
-        return response.choices[0].message.content
+        answer = response.choices[0].message.content
+        answer = re.sub(r'\[\d+\]', '', answer)
+        return answer   
     except Exception as e:
         # Fallback or error handling for Groq API
         return f"Error generating answer from LLM: {str(e)}"
@@ -304,18 +347,14 @@ class QueryRequest(BaseModel):
     top_k: Optional[int] = Field(default=6, ge=1, le=20)
 
 class EvidenceItem(BaseModel):
-    doc_id: int
     question: str
     answer: str
     category: str
-    similarity: float
-    rerank_score: float
 
 class QueryResponse(BaseModel):
     status: str
     question: str
     answer: str
-    evidence: List[EvidenceItem]
     execution_time_sec: float
 
 # ==========================================
@@ -337,40 +376,34 @@ def health_check():
         "model_loaded": "embedder" in rag_resources
     }
 
-@app.post("/predict", response_model=QueryResponse, tags=["RAG Inference"])
-@app.post("/query", response_model=QueryResponse, tags=["RAG Inference"])
-def predict(payload: QueryRequest):
+def _run_rag(payload: QueryRequest) -> QueryResponse:
     start_time = time.time()
-
     try:
-        # Step 1: Vector Search
         candidates = vector_search(payload.question, top_n=RAG_CONFIG.get("retrieve_top_n", 20))
-        
-        # Step 2: Cross Encoder Rerank
         reranked_df = rerank(payload.question, candidates, top_k=payload.top_k)
-        
-        # Step 3: Deduplicate Evidence
         evidence = build_evidence(reranked_df)
-        
-        # Step 4: Generate LLM Answer
         answer = generate_answer(payload.question, payload.user_data or "", evidence)
-        
         elapsed = round(time.time() - start_time, 3)
-
         return QueryResponse(
             status="success",
             question=payload.question,
             answer=answer,
-            evidence=[EvidenceItem(**item) for item in evidence],
             execution_time_sec=elapsed
         )
-
     except Exception as e:
         raise HTTPException(
             status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
             detail=f"An error occurred during inference: {str(e)}"
         )
 
+@app.post("/query", response_model=QueryResponse, tags=["RAG Inference"])
+def query(payload: QueryRequest):
+    return _run_rag(payload)
+
+@app.post("/predict", response_model=QueryResponse, tags=["RAG Inference"])
+def predict(payload: QueryRequest):
+    return _run_rag(payload)
+
 if __name__ == "__main__":
     import uvicorn
     uvicorn.run("app:app", host="127.0.0.1", port=8000, reload=True)