title	emoji	colorFrom	colorTo	sdk	app_port
ChatBotDocker	🐳	purple	gray	docker	7860

ChatBot

A full-stack chatbot which uses a LLM as its core and gradio as UI. Demo: https://huggingface.co/spaces/nblong/chatbot

The chatbot allows searching on the Internet and storing/reading PDF files for Q/A. In other word, it's a RAG chatbot with search tool integrated.

TechStack & Workflow

• Tech:

  • LangChain/LangGraph: Call LLM and define the workflow of agent
  • ChromaDB: Store embedded vector
  • GitHub Action: Use for CI
  • DockerHub and HuggingFace (HF) Space: Use for CD

• Workflow of agent: It is a typical ReAct agent with 2 tools integrated: Searching on Internet (via Tavily) and Searching in vector database (via ChromaDB)

  

• CI/CD flow: On every push to main branch, GitHubAction starts CI process. If CI process is successful, it starts 2 actions parallel

  • Build image and push to DockerHub
  • Push code to HF Space. HF Space utilizes Dockerfile to build image and run container

Dependency stuff

• Since there are a lot dependencies and resolving their version takes lots of time, I use pip-compile to resolve them before pushing everything to GitHub. pip-compile take requirements.in as input and return a requirements.txt. Just specify needed packages in requirements.in (without version) and pip-compile requirements.in

Run locally

• Pull image from DockerHub

• Config file .env as follows to run the image:

  GOOGLE_API_KEY=<key>
  TAVILY_API_KEY=<key>
  PYTHONPATH=/app/src # so you can run code from `/app/src`

• Note that .env file can contain single/double quote but only use for docker compose up (which means that you have to config a compose.yml). Otherwise, you should not use quote since we can run as follows:

  docker run --env-file .env --name ctn_chatbot -p 5555:7860 baolongnguyenmac/chatbot

• In case you want to run via compose.yml, config the compose file as follows:

  services:
  chatbot:
      image: baolongnguyenmac/chatbot
      container_name: ctn_chatbot
      env_file:
          - .env
  
      ports:
          - 5555:7860

  • Then docker compose up

• Demo link at local: localhost:5555

CI using GitHub Action

• Create 2 secrets to store these keys in GitHub Secret. In the workflow file ./.github/workflows/main.yml, specify the environment by

  runs-on: ubuntu-latest
  env:
      GOOGLE_API_KEY: ${{ secrets.GOOGLE_API_KEY }}
      TAVILY_API_KEY: ${{ secrets.TAVILY_API_KEY }}

• Storing secret to GitHub Secret or Hugging Face Secret means that these secrets will be exported to system variable at runtime

CD to DockerHub

• Create a PAT so GitHubAction can push image to DockerHub
• Save this token to GitHub Secret and specify username of DockerHub account
• Setup main.yml (build_and_push_docker)

CD to HF Space

• Create a Space (specify to use Docker)
• Add the content in the file .env to Secret and Variable in the space. These secrets and variables will be exported to be system variable at runtime
• Setup main.yml (push_HF_space)

Read more

https://huggingface.co/learn/cookbook/rag_evaluation https://docs.ragas.io/en/stable/concepts/metrics/available_metrics/