Retrieval QA Agent

The Retrieval QA Agent combines document retrieval with question answering capabilities. It efficiently searches through a knowledge base to find relevant information and generates accurate, context-aware responses while maintaining source traceability.

Retrieval QA Agent workflow and architecture

Configuration Parameters

Required Input Parameters

input: The question or query to be answered
retriever: Document retrieval configuration
- type: Vector store type (e.g., 'faiss', 'pinecone', 'elasticsearch')
- search_type: Search methodology ('similarity', 'mmr', 'hybrid')
- top_k: Number of documents to retrieve

Optional Configuration

llm: Language model configuration (default: GPT-4)
- model_name: Name of the language model
- temperature: Response creativity (0-1)
- max_tokens: Maximum response length
memory: Memory configuration for conversation context
- type: Memory type ('buffer', 'summary', 'conversation')
- max_tokens: Memory context window size
- persistence: Memory storage configuration
return_source_documents: Include source documents in output (default: false)

Output Format

{
  "text": string,  // The generated answer
  "source_documents": [  // Optional, if return_source_documents is true
    {
      "content": string,
      "metadata": {
        "source": string,
        "page": number,
        "relevance_score": number
      }
    }
  ],
  "metadata": {
    "retrieval_time": number,
    "processing_time": number,
    "confidence_score": number
  }
}

Features

Semantic search capabilities
Multi-document reasoning
Context-aware responses
Source attribution
Conversation memory
Configurable retrieval strategies
Answer confidence scoring
Document relevance ranking

Note: The quality of responses depends significantly on the quality and organization of your document store. Regular maintenance and updates of the knowledge base are recommended for optimal performance.

Tip: Use the MMR (Maximal Marginal Relevance) search type when you need to balance relevance with diversity in retrieved documents. Adjust the top_k parameter based on your specific use case and required response comprehensiveness.

Example Usage

const retrievalQA = new RetrievalQA({
  llm: {
    model_name: "gpt-4",
    temperature: 0.7,
    max_tokens: 500
  },
  retriever: {
    type: "faiss",
    search_type: "mmr",
    top_k: 5
  },
  memory: {
    type: "conversation",
    max_tokens: 2000
  },
  return_source_documents: true
});

const response = await retrievalQA.query({
  input: "What are the key features of our product?",
});