| """Grounding prompts for strict context adherence""" |
|
|
|
|
| class GroundingPrompts: |
| """Structured prompts designed for strict context grounding""" |
| |
| @staticmethod |
| def system_prompt() -> str: |
| """System prompt enforcing strict context adherence""" |
| return """You are a helpful technical support assistant for product documentation. |
| |
| CRITICAL INSTRUCTION: You MUST only answer questions using the provided context documents. |
| |
| Rules: |
| 1. ONLY use information from the provided context chunks. |
| 2. If the context doesn't contain the answer, respond: "I don't have information about that in the documentation." |
| 3. Do NOT use external knowledge or make assumptions. |
| 4. Always cite which document(s) your answer comes from. |
| 5. If you're uncertain, ask for clarification or say you need more information. |
| 6. Be concise and clear in your responses. |
| 7. Format citations as: [Source: document_name]""" |
| |
| @staticmethod |
| def build_rag_prompt( |
| query: str, |
| context_chunks: list[str], |
| sources: list[str], |
| ) -> str: |
| """ |
| Build a complete RAG prompt with context. |
| |
| Args: |
| query: User question |
| context_chunks: List of relevant document chunks |
| sources: List of source document names |
| |
| Returns: |
| Complete prompt with context |
| """ |
| context_text = "\n\n".join([ |
| f"[Source: {source}]\n{chunk}" |
| for chunk, source in zip(context_chunks, sources) |
| ]) |
| |
| prompt = f"""Based on the following documentation excerpts, answer the question. |
| |
| DOCUMENTATION CONTEXT: |
| {context_text} |
| |
| QUESTION: {query} |
| |
| ANSWER:""" |
| return prompt |
| |
| @staticmethod |
| def build_verification_prompt( |
| original_query: str, |
| retrieved_chunks: list[str], |
| model_response: str, |
| ) -> str: |
| """ |
| Build a prompt for verifying if response is grounded in context. |
| """ |
| context_text = "\n\n".join(retrieved_chunks) |
| |
| prompt = f"""Verify if the following response is grounded in the provided documentation. |
| |
| DOCUMENTATION: |
| {context_text} |
| |
| ORIGINAL QUESTION: {original_query} |
| |
| MODEL RESPONSE: {model_response} |
| |
| VERIFICATION QUESTIONS: |
| 1. Does the response only use information from the documentation? |
| 2. Are all claims properly sourced? |
| 3. Is there any information that seems to come from outside the documentation? |
| |
| Provide a brief verification result.""" |
| return prompt |
|
|
|
|
| class ResponseBuilder: |
| """Build structured responses with citations""" |
| |
| @staticmethod |
| def build_response( |
| answer: str, |
| sources: list[str], |
| confidence: float = 1.0, |
| ) -> dict: |
| """ |
| Build a structured response with metadata. |
| |
| Args: |
| answer: The generated answer |
| sources: List of source documents |
| confidence: Confidence score (0-1) |
| |
| Returns: |
| Structured response dictionary |
| """ |
| return { |
| "answer": answer, |
| "sources": list(set(sources)), |
| "confidence": confidence, |
| "grounded": True, |
| } |
| |
| @staticmethod |
| def build_fallback_response( |
| query: str, |
| reason: str = "No relevant information found", |
| ) -> dict: |
| """Build a response when no context is available""" |
| return { |
| "answer": f"I don't have information about that in the documentation. {reason}", |
| "sources": [], |
| "confidence": 0.0, |
| "grounded": False, |
| } |
|
|
