Chatbot with RAG

The technologies used are:

• LangChain
• OpenSearch
• Redis

This project consists of two parts:

• a program /src/pdftovector that indexes PDF files under /data to create context for the chatbot
• a program /src/replytouser that takes a question, recontextualizes it based on conversation history saved in Redis, and generates a response from the context

The project is designed to run both autonomously and as a microservice within Lambda or Fargate, so it's already designed to accept input parameters such as consent acceptance, session and user ID.

Console output includes benchmarks to monitor the time the chatbot takes to respond.

Structure

In the root directory you'll find:

• .nvmrc NodeJS version
• .eslint.config.js linter configuration
• .prettierrc formatter configuration
• .husky/ contains git hooks
• docker-compose.yaml container for local development
• package.json dependencies and development scripts
• src/ all project microservices
• data/ directory containing PDF files to be indexed

Essential Steps

1. Run nvm use or manually set NodeJS to the version contained in the .nvmrc file
2. Install development dependencies in the root and for each service install layer dependencies with npm install
3. Populate the .env file
4. Open src/index.js and populate the generateResponse function event with a question
5. Run everything with npm run dev

Local Services

Redis and OpenSearch are available locally as containers, the docker-compose can be managed with npm run docker:start and npm run docker:stop.

Linting and Formatting

eslint and prettier are used to perform linting and format code in an opinionated manner.

Actions can be executed independently with npm run lint and npm run format from the root.

Additionally, through Git Hooks and husky, formatting occurs before every commit and linting before every push. During this phase only staged files are considered, through lint-staged.

Testing

The included test runner is vitest, compatible with Jest.

Tests are run from the root with npm run test.