How chiragandai/summereyes works
Project Type and Purpose
• It uses Streamlit for the frontend UI, Whisper for speech-to-text, LangChain + FAISS for Retrieval-Augmented Generation (RAG), and an LLM (LLaMA 4 via Groq) for summarization.
• Purpose: let users upload PDFs or speak into the app, retrieve relevant context from a local FAISS index, and generate concise summaries.
File Structure
• Top-level directories:
• BackEnd/ – pdf_utils.py (extract text from PDFs)
speechtotext.py (wraps Whisper)
• FrontEnd/ – app.py (Streamlit UI entrypoint)
“we are in a pickle.py” (likely additional UI logic or helpers)
• SummerEyes_local_db/ – index.faiss & index.pkl (persistent FAISS vector store)
• requirements.txt – dependencies
• README.md – project overview
• No separate database server; FAISS files live locally. Deployment is via Streamlit Cloud, inferred from README.
README for Additional Insights
• Workflow described: voice → Whisper → transcript; PDF upload → pdf_utils → raw text → chunking → FAISS embedding → retrieval; pass retrieved context + user input to LLaMA for summarization.
• Tech stack confirmed: Streamlit, Whisper, LLaMA 4, Groq, LangChain, FAISS.