ma2za / docqa-stream

Document QA with FastAPI, Weaviate and Azure OpenAI

DocQA Stream with FastAPI, Weaviate and Azure OpenAI

I noticed that there aren't many examples on how to deploy via docker compose a FastAPI application backed by a self-hosted Weaviate vector store, so here is one.
The interaction with Azure OpenAI happens through langchain.

Quickstart

The API exposes 2 endpoints:

• /files/upload: a POST with a .pdf document to store in a weaviate collection.

• /files/query: a GET with a question to ask based on the uploaded documents. The response will be streamed back to the user.

1) Set the environment

Create a .env in the project root folder in order to set up the environment variables:

WEAVIATE_SERVICE_NAME=weaviate
FASTAPI_PORT=8000
WEAVIATE_PORT=8080
WEAVIATE_COLLECTION = Document
WEAVIATE_DROP_COLLECTION = False
OPENAI_API_TYPE = azure
OPENAI_API_VERSION = 2023-07-01-preview
OPENAI_DEPLOYMENT_NAME=
OPENAI_API_KEY = 
OPENAI_API_BASE = 

If those ports are not in use then you can leave these variables as they are, you just need to set OPENAI_API_KEY, OPENAI_API_BASE and OPENAI_DEPLOYMENT_NAME with the values you can find on the Azure dashboard of your Azure Open AI resource deployment.

2) Docker compose

docker compose up --build

Once it is done, you can check if everything is working fine by making a GET to the following endpoint:

http://localhost:8000/health

If you get an OK message than you are ready to go. You can go the openAPI endpoint and start testing the API:

http://localhost:8000/docs

Run tests:

docker exec fastapi-application poetry run pytest .