hello fix

service/rag_service.py

@@ -39,22 +39,31 @@ vector_store = VectorStoreService(embeddings, documents)
 # -----------------------------
 def generate_answer(question: str, k: int = 3) -> str:
     """
-    Generates a detailed answer for a question using:
-    - RAG (top-k retrieved context from CSV)
-    - TinyLlama LLM for natural language generation
+    Generate an answer strictly from context.
+    If the answer is not present, return the context instead of making up a solution.
     """
 
-    # 1️⃣ Embed the question
     query_vec = embedder.embed([question])[0]
+    top_docs = vector_store.search(query_vec, k=k)
 
-    # 2️⃣ Retrieve top-k context documents
-    top_docs = vector_store.search(query_vec, k=k)  # list of context strings
+    # Deduplicate context
+    seen = set()
+    unique_docs = []
+    for doc in top_docs:
+        if doc not in seen:
+            seen.add(doc)
+            unique_docs.append(doc)
 
-    # 3️⃣ Build a clear instructional prompt
-    context_text = "\n\n".join(top_docs)
+    context_text = "\n\n".join(unique_docs)
+
+    # Build clear, strict prompt
     prompt = f"""
-You are a helpful IT support assistant. Use the context below to provide a detailed, step-by-step troubleshooting guide.
-If the answer is not in the context, say "I don't know".
+You are a helpful IT support assistant.
+
+- ONLY answer based on the context below.
+- DO NOT hallucinate or invent new steps.
+- If the answer is not explicitly present in the context, return the context itself.
+- Keep answers concise and step-by-step if possible.
 
 Context:
 {context_text}
@@ -64,6 +73,4 @@ User question: {question}
 Answer:
 <|assistant|>
 """
-
-    # 4️⃣ Generate response using LLM
     return llm.generate(prompt)