🧠 RAGVerse AI — Chat With Any PDF Book

Upload any PDF book and have a real conversation with it — grounded, cited answers pulled straight from its pages.

Live Demo · Report a Bug · Request a Feature

---

📖 About The Project

RAGVerse AI turns any PDF book into a conversational partner. Drop in a file, and instead of skimming hundreds of pages yourself, you just ask — and get answers pulled directly from the source, with the exact page cited.

Under the hood it's a full Retrieval-Augmented Generation (RAG) pipeline, built end-to-end rather than wrapped around a single prompt: chunking, embeddings, a per-session vector database, MMR-based retrieval, and a strict "answer only from context" prompt so the model doesn't hallucinate content that isn't actually in your book.

This project was built to practice production-style RAG system design — document ingestion, vector search tuning, session-state management, and a fully custom UI — rather than a toy llm.invoke(prompt) demo.

🎯 What It Does

Upload any PDF — a textbook, a novel, a research paper, a manual — and it will:

1. 📄 Load and parse the PDF page by page
2. 🧩 Split it into overlapping chunks for better retrieval accuracy
3. 🧠 Embed every chunk using Mistral's embedding model
4. 🗂️ Build a fresh, isolated vector store for that specific book
5. 💬 Let you chat with it — retrieving the most relevant passages for every question
6. 📚 Show you exactly which page(s) each answer came from

---

🔗 Live Demo

👉 ragverse-ai-garv.streamlit.app

No installation needed — open the link, upload a PDF, and start asking questions.

Repository: github.com/garvkumarsharma/RAGVerse-AI-Project

---

🖼️ Screenshots

Upload Screen	Chat + Sources	Settings Sidebar

---

✨ Key Features

• 📚 Upload any PDF, instantly — no pre-indexing or setup, just drag and drop
• 🧠 True RAG pipeline — chunking → embedding → vector search → grounded generation, not a single prompt trick
• 🔍 MMR retrieval — Maximal Marginal Relevance balances relevance and diversity so answers aren't built from redundant chunks
• 🎛️ Tunable retrieval settings — adjust chunk size, chunk overlap, and number of retrieved chunks live from the sidebar
• 🙅 No hallucinated answers — the model is instructed to say so explicitly when the document doesn't contain the answer
• 📎 Transparent sourcing — every answer comes with an expandable panel showing the exact source passages and page numbers
• 🔄 Per-session isolation — each uploaded book gets its own vector store; switch books anytime without restarting
• 🎨 Custom-designed UI — fully custom dark, gradient-themed Streamlit interface, not default styling

---

🏗️ How It Works — Architecture

                    ┌───────────────────┐
   PDF Upload ────▶ │   PyPDFLoader      │  → Loads and parses the PDF page by page
                    └─────────┬──────────┘
                              ▼
                    ┌───────────────────┐
                    │  Text Splitter     │  → Chunks text (adjustable size / overlap)
                    └─────────┬──────────┘
                              ▼
                    ┌───────────────────┐
                    │ Mistral Embeddings │  → Embeds every chunk into vector space
                    └─────────┬──────────┘
                              ▼
                    ┌───────────────────┐
                    │   Chroma Vector    │  → Fresh, isolated store per uploaded book
                    │      Store         │
                    └─────────┬──────────┘
                              ▼
   User Question ──▶ ┌───────────────────┐
                    │  MMR Retriever      │  → Fetches top-k relevant, diverse chunks
                    └─────────┬──────────┘
                              ▼
                    ┌───────────────────┐
                    │  Context-Bound      │  → "Answer ONLY from this context" prompt
                    │  Prompt Template    │
                    └─────────┬──────────┘
                              ▼
                    ┌───────────────────┐
                    │  Mistral Chat LLM  │  → Generates the grounded answer
                    └─────────┬──────────┘
                              ▼
              Answer + Cited Source Passages (in-app)

---

🛠️ Tech Stack

Layer	Technology
LLM	Mistral AI (mistral-medium-3-5) via langchain-mistralai
Embeddings	Mistral AI Embeddings (langchain-mistralai)
Vector Store	ChromaDB via langchain-chroma
Orchestration	LangChain (document loaders, text splitters, prompt templates)
Document Parsing	pypdf / PyPDFLoader
UI / Frontend	Streamlit (custom CSS, no default theme)
Deployment	Streamlit Community Cloud
Language	Python 3.10+

---

📂 Project Structure

RAGVerse-AI-Project/
├── app.py               # Streamlit UI — main entry point for deployment
├── create_db.py         # Standalone script for building a persistent vector store from a fixed PDF (reference/legacy)
├── main.py               # CLI-based query loop against a persistent vector store (reference/legacy)
├── requirements.txt      # Python dependencies
├── config.toml           # Streamlit configuration (theme/server settings)
├── document_loaders/      # Sample/local PDFs used for testing (gitignored)
├── .gitignore             # Excludes .env, .venv, chroma_db/, *.sqlite3
└── README.md

Note: app.py builds a fresh, per-upload vector store at runtime — it does not depend on create_db.py or main.py. Those two are kept in the repo as standalone reference scripts for building/querying a persistent, pre-indexed vector store outside the web UI.

---

🚀 Getting Started — Run It Locally

Prerequisites

• Python 3.10 or higher
• A Mistral AI API key

Installation

1. Clone the repository

git clone https://github.com/garvkumarsharma/RAGVerse-AI-Project.git
cd RAGVerse-AI-Project

2. Create and activate a virtual environment

python -m venv .venv

# Windows
.venv\Scripts\activate

# macOS/Linux
source .venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. Set up environment variables

Create a .env file in the project root:

MISTRAL_API_KEY=your_mistral_api_key_here

⚠️ Never commit your .env file. It's already excluded via .gitignore.

5. Run the app

streamlit run app.py

The app will open at http://localhost:8501. Upload a PDF, click Process Book, and start chatting.

---

☁️ Deployment

This project is deployed on Streamlit Community Cloud, connected directly to this GitHub repository.

Live app: ragverse-ai-garv.streamlit.app

If you'd like to deploy your own fork:

1. Push your fork to GitHub
2. Go to share.streamlit.io → Create app
3. Point it at your repo, branch main, main file app.py
4. Under Advanced settings → Secrets, add:

   MISTRAL_API_KEY = "your_key_here"

5. Deploy 🚀

---

🧭 Roadmap / Future Improvements

• Support multiple document formats (.docx, .txt, .epub)
• Multi-document chat (query across several uploaded books at once)
• Persistent chat history across sessions
• Support for additional LLM providers (OpenAI, Gemini) via a model switcher
• Downloadable chat transcript / summary export

---

👤 Author

Garv Kumar Sharma

• GitHub: @garvkumarsharma
• LinkedIn: linkedin.com/in/garv-kumar-sharma
• Email: garvdhiman2004@gmail.com

---

📄 License

This project is open source and available under the MIT License.

---

If you found this project interesting, consider giving it a ⭐ on GitHub!