Skip to content

Latest commit

 

History

History
32 lines (24 loc) · 1.49 KB

README.md

File metadata and controls

32 lines (24 loc) · 1.49 KB
Raw

Funny RAG

Screenshot

Overview

This project is only for educational purpose, it aims to teach myself about the fundemental workflow of RAG (Retrieval-Augmented Generation). The app is powered by FastAPI with Jinja2 template engine. Besides that, for autocomplete system, the trie data structured is also utilized (using submodule named trie).

Features

  • Autocomplete Functionality: Utilizes a trie data structure for efficient autocomplete suggestions based on LLM generated questions.
  • RAG System: Leverages the concept of Retrieval-Augmented Generation for enhancing information retrieval and response generation (In progress).
  • LLM Integration: Uses a Large Language Model to analyze data, identify trends, and generate relevant questions. [Looking options for utilizing open-source LLMs with faster inference methods (tvm, llama.cpp, wasm)].
  • LLM Used: -- Mistral-7B API provided by together.ai
    -- GPT-3.5 powered by OpenAI

Installation

Clone:
https://github.com/author31/funny-rag.git

Install packages
pip install -r requirements.txt

Run:
uvicorn main:app --port 8000 --host 0.0.0.0 --reload

TODO

  • Crawl the corresponding post linked urls to: -- To generate comprehensive questions for clustered posts. -- To generate insightful answer.
  • Build connections between LLMs and data source
  • Deploy to AWS, only using free-tier service.