Tech stack:
- local DB postgresql
- vector DB Pinecone
- ORM Prisma
- OpenAPI (Embedding/Completion)
- Langchain integration (HNSWLib, OpenAI API, Buffer memory)
curl -X POST \
-H "Content-Type: application/json" \
-d '{
"message": "Do you have a question?"
}' \
http://localhost:2137/assistant/conversation
Repository includes some experiments with the current state of langchain. There is an endpoint (docs/upload) which allows to upload and store embeddings (generated from uploaded files) within in-memory vector db (HNSWLib). Then you can make request to query endpoint (docs/query) asking about any information relevant to uploaded docs. Relevant data is gathered through similarity search.
The response is generated with the help of OpenAI GPT-3.5-turbo.