Research Paper Explainer using RAG + AI

This repo contains all code for running a paper explained by AI given a user input. This was made at its simplest level with no UI, just running commands.

Purpose

Demonstration of understanding

Steps to run:

1. Run download_model.sh to first download 'Meta-Llama-3-8B-Instruct.Q2_K' locally given that you created a HuggingFace account. This will also create a models/llama3 nested file.
2. Run file_manager.py to insert your target research paper. Only one file can be read at a given time. This will create a data/papers nested file.
3. Run run_indexer.py to embed the target research paper. This will create a vector store which will be using by the LLM later on. run_indexer.py uses subcomponents from src/indexer/subcomponents Models/Algo employed for this task:
   • Langchain Recursive Text Splitter
   • FAISS
   • Sentence Transformer: all-MiniLM-L6-v2
4. Run main.py which orchestrates everything together

Do keep in mind that this serves as a POC for future project as many improvements are planned to be done.

File Contents

src contains main working directories: indexer, retriever, llm

indexer Contains components used to build indexer: subcomponent directory, paper_indexer.py which puts all subcomponents together

retriever Contains vec_loader and vec_retriever which are used for model RAG

llm Contains components loading the LLM and what it output.

• Loads the model load_local_model.py
• Reads paper and answers with paper_reader.py
• prompt.py Main prompt for confining the LLM
• citator.py provides sources of where the information was found