paper2summary

A resource-efficient RAG engine for scientific paper question answering

Quick Summary

  • Developed a RAG engine for scientific paper QA with accurate source citations
  • Enhanced Llama-3.2-1B for paper summarization via LoRA fine-tuning (7% trainable parameters), achieving 51% higher ROUGE-2 and 13% ROUGE-L scores
  • Deployed a hybrid text/vector retrieval system featuring LLM reranking and citation visualization on laptop hardware

Tools: PEFT, Transformers, Weights & Biases, Kotaemon, ChromaDB

Video Demo

For more detailed project information, please visit the GitHub repo: https://github.com/gabe-zhang/paper2summary