I'm happy to help you use files as a knowledge base in LiveKit! One effective approach is to implement a Retrieval-Augmented Generation (RAG) system. This means you'll be processing text and structured data from your files, creating a searchable vector database, and then integrating it into a voice or chat agent. To make things clearer, here’s a step-by-step code example to guide you through building a RAG-enabled knowledge base: 1. Preprocessing PDF Files to Text (Client-Side Example):

import * as pdfjs from 'pdfjs-dist';
import { TextItem } from 'pdfjs-dist/types/src/display/api';

async function convertPDFToText(file: File): Promise<string> {
  const buffer = await file.arrayBuffer();
  const pdf = await pdfjs.getDocument({ data: buffer }).promise;
  const numPages = pdf.numPages;
  const textContent: string[] = [];

  for (let i = 1; i <= numPages; i++) {
    const page = await pdf.getPage(i);
    const text = await page.getTextContent();
    const pageText = text.items.map(item => (item as TextItem).str).join(' ');
    textContent.push(pageText);
  }

  return textContent.join('\n');
}

2. Build RAG Database from Text Files:

import asyncio
from rag_db_builder import RAGBuilder

async def main():
    await RAGBuilder.create_from_file(
        file_path="knowledge_base.txt",
        index_path="data",
        data_path="paragraphs.pkl",
        embeddings_dimension=1536,
        embeddings_model="text-embedding-3-small"
    )

if __name__ == "__main__":
    asyncio.run(main())

3. RAG-Enabled Agent Using LlamaIndex:

from llama_index.core import SimpleDirectoryReader, VectorStoreIndex

documents = SimpleDirectoryReader("data").load_data()
index = VectorStoreIndex.from_documents(documents)

async def query_info(query: str) -> str:
    query_engine = index.as_query_engine(use_async=True)
    response = await query_engine.aquery(query)
    return str(response)

If you have any questions or need more details on file ingestion or troubleshooting, please don’t hesitate to reach out. I’m here to support you every step of the way! Sources: livekit/livekit_composite | DeepWiki | examples/retrieval/retrieval.cpp | rag/rag_db_builder.py | examples/voice_agents/llamaindex-rag/retrieval.py