Upd RAG for sym-diagnosis

app.py

@@ -94,6 +94,9 @@ except Exception as e:
     logger.error(f"❌ Model Loading Failed: {e}")
     exit(1)
 
+# Cache in-memory vectors (optional — useful for <10k rows)
+SYMPTOM_VECTORS = None
+SYMPTOM_DOCS = None
 
 # ✅ Setup MongoDB Connection
 # QA data
@@ -104,6 +107,9 @@ qa_collection = db["qa_data"]
 iclient = MongoClient(index_uri)
 idb = iclient["MedicalChatbotDB"]
 index_collection = idb["faiss_index_files"]
+# Symptom Diagnosis data
+symptom_client = MongoClient(mongo_uri) 
+symptom_col = symptom_client["MedicalChatbotDB"]["symptom_diagnosis"]
 
 # ✅ Load FAISS Index (Lazy Load)
 import gridfs
@@ -142,6 +148,26 @@ def retrieve_medical_info(query, k=5, min_sim=0.6): # Min similarity between que
             results.append(doc.get("Doctor", "No answer available."))
     return results if results else ["No relevant medical entries found."]
 
+# ✅ Retrieve Sym-Dia Info (4962 scenario)
+def retrieve_diagnosis_from_symptoms(symptom_text, top_k=5, min_sim=0.4):
+    global SYMPTOM_VECTORS, SYMPTOM_DOCS
+    # Lazy load
+    if SYMPTOM_VECTORS is None:
+        all_docs = list(symptom_col.find({}, {"embedding": 1, "answer": 1, "question": 1}))
+        SYMPTOM_DOCS = all_docs
+        SYMPTOM_VECTORS = np.array([doc["embedding"] for doc in all_docs], dtype=np.float32)
+    # Embed input
+    qvec = embedding_model.encode(symptom_text, convert_to_numpy=True)
+    qvec = qvec / (np.linalg.norm(qvec) + 1e-9)
+    # Similarity compute
+    sims = SYMPTOM_VECTORS @ qvec  # cosine
+    sorted_idx = np.argsort(sims)[-top_k:][::-1]
+    # Final
+    return [
+        SYMPTOM_DOCS[i]["answer"]
+        for i in sorted_idx
+        if sims[i] >= min_sim
+    ]
 
 # ✅ Gemini Flash API Call
 def gemini_flash_completion(prompt, model, temperature=0.7):
@@ -161,8 +187,11 @@ class RAGMedicalChatbot:
 
     def chat(self, user_id: str, user_query: str, lang: str = "EN") -> str:
         # 1. Fetch knowledge
+        ## a. KB for generic QA retrieval
         retrieved_info = self.retrieve(user_query)
         knowledge_base = "\n".join(retrieved_info)
+        ## b. Diagnosis RAG from symptom query
+        diagnosis_guides = retrieve_diagnosis_from_symptoms(user_query)  # smart matcher
 
         # 2. Use relevant chunks from short-term memory FAISS index (nearest 3 chunks)
         context = memory.get_relevant_chunks(user_id, user_query, top_k=3)
@@ -177,6 +206,9 @@ class RAGMedicalChatbot:
         # Load up guideline
         if knowledge_base:
             parts.append(f"Medical knowledge (256,916 medical scenario): {knowledge_base}")
+        # Symptom-Diagnosis prediction RAG
+        if diagnosis_guides:
+            parts.append("Symptom-based diagnosis guidance:\n" + "\n".join(diagnosis_guides))
         parts.append(f"Question: {user_query}")
         parts.append(f"Language: {lang}")
         prompt = "\n\n".join(parts)