Update rag_dspy.py

rag_dspy.py

@@ -12,7 +12,7 @@ load_dotenv()
 # DSPy setup
 lm = dspy.LM("gpt-4", max_tokens=512,api_key=os.environ.get("OPENAI_API_KEY"))
 client = QdrantClient(url=os.environ.get("QDRANT_CLOUD_URL"), api_key=os.environ.get("QDRANT_API_KEY"))
-collection_name = "indexed_medical_chat_bot"
+collection_name = "medical_chat_bot"
 rm = QdrantRM(
     qdrant_collection_name=collection_name, 
     qdrant_client=client, 
@@ -24,7 +24,7 @@ dspy.settings.configure(lm=lm, rm=rm)
 
 # Manual reranker using ColBERT multivector field
 # Manual reranker using Qdrant’s native prefetch + ColBERT query
-def rerank_with_colbert(query_text, year, specialty):
+def rerank_with_colbert(query_text, min_year, max_year, specialty):
     from fastembed import TextEmbedding, LateInteractionTextEmbedding
 
     # Encode query once with both models
@@ -48,8 +48,8 @@ def rerank_with_colbert(query_text, year, specialty):
         query_filter=Filter(
             must=[
                 FieldCondition(key="specialty", match=MatchValue(value=specialty)),
-                FieldCondition(key="year", match=MatchValue(value=year))
-                ]
+                FieldCondition(key="year",range=models.Range(gt=None,gte=min_year,lt=None,lte=max_year))
+            ]
 
         )
     )
@@ -66,7 +66,8 @@ def rerank_with_colbert(query_text, year, specialty):
 class MedicalAnswer(dspy.Signature):
     question = dspy.InputField(desc="The medical question to answer")
     is_medical = dspy.OutputField(desc="Answer 'Yes' if the question is medical, otherwise 'No'")
-    year = dspy.InputField(desc="The year of the medical paper")
+    min_year = dspy.InputField(desc="The minimum year of the medical paper")
+    max_year = dspy.InputField(desc="The maximum year of the medical paper")
     specialty = dspy.InputField(desc="The specialty of the medical paper")
     context = dspy.OutputField(desc="The answer to the medical question")
     final_answer = dspy.OutputField(desc="The answer to the medical question")
@@ -87,16 +88,17 @@ class MedicalRAG(dspy.Module):
         super().__init__()
         self.guardrail = MedicalGuardrail()
 
-    def forward(self, question, year, specialty):
+    def forward(self, question, min_year, max_year, specialty):
         if not self.guardrail.forward(question):
             class DummyResult:
                 final_answer = "Sorry, I can only answer medical questions. Please ask a question related to medicine or healthcare."
             return DummyResult()
-        reranked_docs = rerank_with_colbert(question, year, specialty)
+        reranked_docs = rerank_with_colbert(question, min_year, max_year, specialty)
         context_str = "\n".join(reranked_docs)
         return dspy.ChainOfThought(MedicalAnswer)(
             question=question,
-            year=year,
+            min_year=min_year,
+            max_year=max_year,
             specialty=specialty,
             context=context_str
         )