Following this to create a RAG without langchain.

This project implements streaming and RAG over pdf, and gives answer according to the pdf provided. Implemented using:

• FastAPI
• Postgres
• Ollama
• SvelteKit
• Docker

I wanted to create a app that didn't use langchain, although using it would have been quite easy. Since langchain is a framework and hides away a lot of complexities behind its abstractions customizing it can be a pain.

Here is how it works in normal chat (It's a little slow in the first response):

Here is how it works in Pdf chat, I used the famous overleaf software engineer resume:

To run this just run

docker compose up --build

it will start front end and backend on the same host
Frontend: http://localhost:5173/
Backend: http://localhost:8000/docs (FastAPI)

• Response streaming (typing words as soon as it is recived from server)
• Pdf Chat (upload a pdf and ask questions about it)
• Normal Chat (you can chat normally, Currently using llama3 model but can be customized)

• Preserving context over chats (after uploading pdf, its context should be cached or saved)
• Caching responses to produce better results
• Make UI more good looking
• Add support for images, videos etc.
• Fast responses