Skip to content

README.md

Latest commit

 

History

History
88 lines (58 loc) · 2.68 KB

README.md

File metadata and controls

88 lines (58 loc) · 2.68 KB

🧠 GenAI & RAG Pipeline

This module implements the core GenAI capabilities for Study Mate, leveraging Retrieval-Augmented Generation (RAG) and LangChain to enable document-aware responses, summaries, flashcards, and quizzes.

🔍 What It Does

When a user uploads a .pdf or .txt document, the system:

  • ✅ Parses and splits the content into meaningful chunks
  • 🧠 Embeds content using HuggingFace (MiniLM) and stores in Weaviate
  • 💬 Supports document-specific chat using RAG
  • ✍️ Generates:
    • Structured summaries (Markdown)
    • Flashcards (difficulty-tagged)
    • Quizzes (MCQ and short-answer)

📁 Key Files

File Description
llm.py Manages all GenAI functionality: chat, summarization, flashcards, quiz generation
rag.py Handles ingestion, chunking, metadata, vector embedding, and retrieval via Weaviate
chains.py Defines custom LangChain chains to generate structured outputs (flashcards, quizzes)

🧠 Architecture Overview

  • 🔗 LangChain for LLM orchestration
  • 📚 Weaviate stores two types of chunks:
    • RAGChunksIndex: Vectorized, small chunks for semantic search
    • GenerationChunksIndex: Larger, plain text chunks for generative tasks (e.g., summaries)
  • 🤗 HuggingFace MiniLM used for embeddings
  • 🌐 Open WebUI-compatible API (LLaMA 3) for LLM calls

🗂 Workflow

  1. Load document
    → via StudyLLM.load_document(doc_name, path, user_id)

  2. Ingest chunks
    → Embedded RAG chunks go to RAGChunksIndex
    → Generation chunks go to GenerationChunksIndex

  3. Chat
    → Queries are filtered by user_id and optionally doc_name
    → Top-k relevant chunks retrieved from Weaviate and passed to the LLM

  4. Summarize / Flashcards / Quiz
    → Uses larger plain-text chunks stored per document
    → LangChain’s map-reduce pattern parallelized to generate structured output

🧪 Available Features

Feature Method
Load Document load_document(doc_name, path, user_id)
RAG Chat prompt(prompt, user_id)
Summarize summarize(document_name, user_id)
Flashcards generate_flashcards(document_name, user_id)
Quiz generate_quiz(document_name, user_id)
Cleanup (see below) cleanup()

📌 Notes

  • Only .pdf and .txt documents are supported.

📦 Dependencies

  • langchain
  • langchain-openai
  • langchain-huggingface
  • langchain-community
  • weaviate-client
  • PyMuPDF, dotenv, asyncio, etc.

🧹 Cleanup

Call StudyLLM.cleanup() to close the Weaviate client connection properly.

📄 License

MIT — see LICENSE