Conversational RAG Chatbot 🤖

An App for Exploring Retrieval-Augmented Generation on Your Own Docs

RAG architecture.

🚀 Project Overview

This project is a modern, privacy-friendly chatbot app that demonstrates genuine Retrieval-Augmented Generation (RAG) in the browser. Users can securely upload their own .txt, .pdf, or .docx documents and have a conversation with an AI assistant that grounds its answers entirely in the user’s files, no preloaded data, no external search.

Key features include:

• True document-based RAG chat: every answer is sourced from your uploaded references.
• Instant “Demo Mode” lets anyone try the chatbot using a bundled demo file if they prefer not to upload.
• Conversely, each session is fully private, no file, chat, or memory leaks between users.
• Seamless follow-up Q&A: the app converts every follow-up into a standalone question, letting you interact naturally.
• All processing and chat state are wiped with a single click for a refreshed, secure experience.

---

💡 How It Works (Tech Stack)

• Frontend/UI: Streamlit chat UI for smooth user interaction.
• Document Parsing: LangChain community loaders for TXT, PDF, DOCX.
• Chunking & Embedding: Fast local embedding via sentence-transformers/all-MiniLM-L6-v2.
• Vector Storage: In-memory ChromaDB per user session for similarity search (never persisted on disk, privacy by design).
• Language Model (Q&A): Responses powered by a state-of-the-art, cloud-hosted LLM (e.g. MoonshotAI Kimi-K2-Instruct) via the Hugging Face Inference API.
• Session/state: Strict isolation per user/session, no data cross-contamination, guaranteed.

---

✅ Demo Use-Cases

• Showcases RAG for recruiters/colleagues: upload a resume, handbook, or research article and chat with your own knowledge!
• Compare demo mode to real document upload, see the assistant adapt instantly.
• Safe, ephemeral environment for testing generative-AI techniques with privacy by default.

---

🚧 Demo Limitations & Notes

• Speed: Model loading and first LLM answer may take 10–30s, very normal for open RAG demos using real embeddings and cloud APIs.
• File Limits: Files must be .txt, .pdf, or .docx (max 2MB each, up to 30,000 total characters per session).
• Session Lifetime: Sessions are wiped on page reload, reset, or user tab close, for privacy, not for persistence.
• No “real” cross-user chat: Each user is isolated, great for privacy, but not for multiplayer chatrooms.
• Cloud API: All LLM completions are performed using Hugging Face’s inference API as configured; usage is subject to your account’s quota.

---

📦 Installation & Running Locally

1. Create a virtual environment:
   python -m venv doc_chat
2. Activate the virtual environment :
   .\doc_chat\Scripts\activate on Windows
   source .doc_chat/bin/activate on Linux/macOS
3. Run the following command in the directory:
   cd rag-document-chatbot
4. Install the required dependencies:
   pip install -r requirements.txt
5. Start the app:
   streamlit run app.py
6. The web UI may take 10-30 seconds to render. Once it does, upload the document (within the constrains) you want to query.
7. Talk to your document: ask questions and get answers.

---

Screenshot(s)

Landing page UI.

---