🤖 TellMe.ai — AI-Powered Multi-Document Reader & Voice Assistant

M.Sc. Computer Science Final Year Project — Ramniranjan Jhunjhunwala College, Mumbai (2025–2026)

A fully offline, CPU-friendly RAG (Retrieval-Augmented Generation) system that reads PDF documents, answers questions using a local LLM, and supports voice interaction — no GPU, no cloud, no paid API.

📸 Screenshots

Dark Theme	Light Theme

Answer with Citations	Split View — PDF + Chat

---

✨ Key Features

• 📄 Multi-Document Support — Upload multiple PDFs per session, query across all of them
• 🧠 RAG Pipeline — FAISS vector search + phi3:mini for document-grounded answers
• 🚫 Anti-Hallucination — Strict 4-rule prompt: LLM answers ONLY from uploaded documents
• 🎙️ Voice In + Voice Out — Speak questions, hear answers via Web Speech API
• ⚡ 3 Speed Modes — Quick (150 tokens), Standard (350), Deep (700)
• 📑 Split View — View PDF and chat side by side
• 💾 Persistent Sessions — Sessions saved to disk, resume after restart
• 📤 Export to PDF — Download full Q&A conversation as PDF
• 🌙 Dark/Light Theme — Toggle anytime

---

🏛️ Architecture — RAG Pipeline

PDF Upload → PyMuPDF text extraction → 500-word chunks (50-word overlap)
    → all-MiniLM-L6-v2 embeddings (384-dim) → FAISS IndexFlatL2

User Question → same embedding model → FAISS Top-K retrieval
    → Strict anti-hallucination prompt → phi3:mini via Ollama (offline)
    → SSE token streaming → React frontend → Source citations

---

🛠️ Tech Stack

Layer	Technology	Purpose
LLM	phi3:mini (3.8B) via Ollama	Local offline inference
Embeddings	all-MiniLM-L6-v2	384-dim semantic vectors
Vector Search	FAISS IndexFlatL2	CPU-based similarity search
PDF Parsing	PyMuPDF (fitz)	Page-by-page text extraction
Backend	FastAPI + Python	REST API + SSE streaming
Frontend	React 18.3 + Vite 5.4	Single-page application
Voice	Web Speech API + pyttsx3	Speech input and output
Export	jsPDF 4.2	Chat-to-PDF export

Hardware: Intel Core i3+ · 8GB RAM · No GPU required · Fully offline after setup

---

🚀 Installation & Setup

Prerequisites

• Python 3.10+
• Node.js 16+
• Ollama installed

1. Clone the Repository

git clone https://github.com/Harh2646/TellMe.ai.git
cd TellMe.ai

2. Backend Setup

python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt
ollama pull phi3:mini

3. Download Embedding Model (one-time, ~90MB)

from sentence_transformers import SentenceTransformer
model = SentenceTransformer("all-MiniLM-L6-v2")
model.save("./models/all-MiniLM-L6-v2")

4. Frontend Setup

cd frontend
npm install

---

▶️ Running the App

Terminal 1 — Start Ollama:

ollama run phi3:mini

Terminal 2 — Start Backend:

python backend/main.py

Terminal 3 — Start Frontend:

cd frontend
npm run dev

Open browser → http://localhost:3000

---

📁 Project Structure

TellMe.ai/
├── backend/
│   └── main.py
├── frontend/
│   ├── src/
│   │   ├── App.jsx
│   │   └── main.jsx
│   ├── index.html
│   ├── package.json
│   └── vite.config.js
├── docs/
│   └── screenshots/
├── requirements.txt
└── README.md

---

🔌 API Endpoints

Method	Endpoint	Description
POST	/sessions	Create new session
GET	/sessions	List all sessions
DELETE	/sessions/{id}	Delete session
POST	/sessions/{id}/upload	Upload PDF
POST	/sessions/{id}/query_stream	Ask question (SSE streaming)
GET	/sessions/{id}/summary	Summarize document
GET	/sessions/{id}/history	Get chat history

---

👨‍💻 Author

Harsh Sanjay Singh M.Sc. Computer Science — Ramniranjan Jhunjhunwala College, Mumbai

• 📧 harshsingh2646@gmail.com
• 💼 LinkedIn
• 🐙 GitHub

---

📄 License

MIT License — see LICENSE file for details.