francodegio/immigration-agent
Australia's Immigration Agent ๐ฆ๐จ
This project was created by Franco De Giovannini
TL;DR
This is a chatbot with context knowledge about Australia's Department of Home Affairs' website. It uses ChatGPT under the hood for quicker response times.
About
Conversational RAG agent using Langchain, FAISS, OpenAI, Streamlit and Docker.
There's also an implementation using llama-cpp-python and faiss-gpu that can be found on llama-cpp branch. Async implementation of endpoints is still required, as well as paralell processing. Performance is poor due to the complexity of the RAG chain.
How to use with proprietary data
1. Create a `.secrets` file inside the `./scripts` directory
DOCKER_IMAGE_NAME=immi-agent-base
DOCKER_TAG=latest
OPENAI_API_KEY='your-openai-api-key'
DOCKERHUB_USERNAME=your-username
DOCKERHUB_PASSWORD=your-token2. Replace `./data` with your own 3. Change dir to `./scripts` and run `build.sh`
$ cd ./scripts
$ sh build.sh3b. [Optional] Build locally to test
$ source .secrets
$ cd ..
$ docker build --build-arg OPENAI_API_KEY=$OPENAI_API_KEY -t $DOCKERHUB_USERNAME/$DOCKER_IMAGE_NAME . 4. Create a secret in your ๐ค space with name `OPENAI_API_KEY` 5. Push to your ๐ค space
Visit the following links more information:
