RAG Service¶

Overview¶

The RagService implements a multi-step retrieval-augmented generation pipeline that connects language models with document context to provide context-aware AI responses.

Dependencies¶

LangchainService: For language model interaction
PineconeService: For vector database operations

Key Methods¶

initializeQA¶

public async initializeQA(
  llm: BaseChatModel,
  db: PineconeStore,
  k: number,
  type?: string,
  themesList?: string[]
)

Initializes the RAG pipeline with: - Question-condensing chain to rewrite user queries - Document retrieval from vector store - Answer generation with document context

Parameters: - llm: Language model to use - db: Pinecone vector store - k: Number of documents to retrieve - type: Optional response type - themesList: Optional themes for filtering

getAnswer¶

async getAnswer(
  vectorStore: PineconeStore,
  query: string,
  lengthType: 'short' | 'long' = 'short',
  themes: string[] = [],
  language: string = 'english-uk'
): Promise<string>

Processes a query and returns a context-aware answer.

Parameters: - vectorStore: Pinecone vector store - query: User question - lengthType: Response length preference - themes: Optional themes for filtering - language: Response language

Returns: - String containing the generated answer

addDocumentToVectorStore¶

async addDocumentToVectorStore(
  documentBase64: string,
  userId: string
): Promise<void>

Adds a document to the vector store: - Converts base64 document to text - Processes and adds document to vector store

Parameters: - documentBase64: Base64-encoded document - userId: User identifier