Retrieval QA Agent

Required Parameters

• Language Model: The AI model for processing
• Knowledge Base: Document collection source
• Retriever Type: Document retrieval method
• Index Type: Document indexing method
• Embedding Model: Text embedding model

Optional Configuration

• Retrieval Options: Search settings
  • topK: Number of documents
  • threshold: Similarity threshold
  • maxTokens: Token limit
• Processing: Processing options
  • chunkSize: Document chunk size
  • overlap: Chunk overlap
  • preprocessing: Text preprocessing
• Caching: Cache settings
  • enabled: Enable caching
  • ttl: Cache lifetime
  • maxSize: Cache size limit

Output Format

{
  "result": {
    "answer": {
      "text": string,
      "confidence": number
    },
    "context": {
      "documents": array,
      "relevance": array
    },
    "metadata": {
      "processingTime": string,
      "sourceCount": number,
      "tokenCount": number
    }
  }
}

Example Usage

const retrievalQAAgent = new RetrievalQAAgent({
  languageModel: "gpt-4",
  knowledgeBase: {
    source: "documents/",
    format: ["pdf", "txt", "md"]
  },
  retrieverType: "hybrid",
  indexType: "faiss",
  embeddingModel: "sentence-transformers/all-mpnet-base-v2",
  retrievalOptions: {
    topK: 5,
    threshold: 0.7,
    maxTokens: 1000
  },
  processing: {
    chunkSize: 500,
    overlap: 50,
    preprocessing: {
      removeStopwords: true,
      lowercase: true,
      stemming: true
    }
  },
  caching: {
    enabled: true,
    ttl: 3600,
    maxSize: "1GB"
  }
});

const result = await retrievalQAAgent.process({
  input: "What are the key features of our product?"
});