Tensor Chat (v3)

Tensor Chat v3 is a fast, modern AI chatbot powered by LLaMA 3.3 70B via Groq, with LangChain memory, Document Q&A capabilities, multi-user session support, a Flask backend, and a beautiful React + Tailwind frontend.

---

Features – v3

• Real-time conversational AI
• Powered by LLaMA 3.3-70B via Groq API
• Maintains memory of last 10 messages (LangChain buffer)
• Document Q&A System - Upload PDFs and ask questions about them
• Semantic Search - HuggingFace embeddings with vector similarity search
• In-Memory Vector Store - Fast document retrieval using DocArray
• Modern responsive chat UI with drag-and-drop PDF upload
• Multi-user session support with PDF-specific sessions
• Avatar icons, typing indicators, and visual PDF status
• Hybrid Chat Mode - Seamlessly switches between general chat and document Q&A

Version 1 (v1) – a Minimum Viable Product (MVP) for the core chat system.

• Streaming responses
• Persistent chat history

Version 2 (v2) – Added multi-user session support, avatar icons and typing indicators.

• Multi-user session support
• Avatar icons and typing indicators

Version 3 (v3) (Active)

• Document Q&A with PDF upload functionality
• Semantic search using HuggingFace embeddings (sentence-transformers/all-MiniLM-L6-v2)
• In-memory vector stores with DocArrayInMemorySearch
• Context-aware answers from uploaded documents
• Intelligent routing between general chat and document Q&A
• Enhanced UI with drag-and-drop PDF upload and visual indicators

Future Roadmap

• WebSocket Integration - Real-time bidirectional communication for instant messaging
• Language Translation Chains - Multi-language support with automatic translation
• Streaming Responses - Real-time token streaming for faster user experience
• Persistent Chat History - Database storage for conversation continuity
• Multiple Document Support - Handle multiple PDFs simultaneously
• Advanced Document Types - Support for Word, Excel, PowerPoint, and more
• Voice Integration - Speech-to-text and text-to-speech capabilities
• Custom Model Support - Integration with other LLM providers

---

Tech Stack

Layer	Technology
Frontend	React, Tailwind CSS, Vite
Backend	Flask, LangChain
LLM	LLaMA 3.3-70B via Groq API
Document AI	LangChain Document Loaders
Embeddings	HuggingFace Transformers
Vector Store	DocArrayInMemorySearch
PDF Processing	PyPDFLoader, pdfminer.six

Document Q&A Features

Semantic Search

• Embedding Model: sentence-transformers/all-MiniLM-L6-v2
• Vector Similarity: Cosine similarity for document chunks
• Context Retrieval: Relevant document sections for each query

Document Processing

• Supported Format: PDF files (up to 16MB)
• Text Extraction: PyPDFLoader with robust parsing
• Chunking Strategy: Automatic text splitting for optimal retrieval

Vector Storage

• In-Memory Store: DocArrayInMemorySearch for fast retrieval
• Session-Based: Separate vector stores per user session
• Real-Time: Instant document processing and indexing

Smart Routing

• Context Detection: Automatically uses document Q&A when PDF is loaded
• Graceful Fallback: Falls back to general chat if document Q&A fails
• Session Persistence: Remembers document state across page refreshes

---

Quick Start

Prerequisites

• Python 3.9+
• Node.js 16+
• Groq API Key (Get one here)

1. Clone & Setup Backend

git clone https://github.com/vishwajayawickrama/Tensor-Chat.git
cd Tensor-Chat

# One command setup (creates venv + installs requirements)
make setup

# Set up environment variables
cp .env.example .env
# Edit .env and add your GROQ_API_KEY

2. Setup Frontend

cd UI
npm install

3. Run the Application

# Terminal 1: Start Flask backend
make run

# Terminal 2: Start React frontend
cd UI && npm run dev

Visit http://localhost:5173

---

Usage Guide

General Chat

• Simply type your message and press Enter
• The AI maintains context of your last 10 messages
• Perfect for general questions, coding help, explanations, etc.

Document Q&A Mode

1. Upload PDF:

   • Drag & drop a PDF onto the upload zone, or
   • Click "Choose PDF File" button, or
   • Use the paperclip button in chat

2. Ask Questions:

   • Once uploaded, ask specific questions about your document
   • Examples: "What is the main topic?", "Summarize the key points", "What does section 3 say about...?"

3. Smart Context:

   • The AI will answer based on your document content
   • If it can't find relevant information, it will tell you clearly
   • Falls back to general knowledge if needed

Hybrid Experience

• With PDF loaded: Document-specific questions get answered from your PDF
• General questions: Still works normally even with PDF loaded
• Session management: Each browser session maintains its own PDF and chat history

---

Project Structure

Tensor-Chat/
├── UI/                             # React Frontend
│   ├── src/App.jsx                 # Main chat interface with PDF upload
│   ├── package.json                # Frontend dependencies
│   └── ...
├── uploads/                        # PDF storage (auto-created)
├── app.py                          # Flask backend with API endpoints
├── session_chain.py               # General chat conversation chains
├── document_qa_chain.py           # Document Q&A processing
├── requirements.txt               # Python dependencies (300+ packages)
├── Makefile                       # Development commands
├── .env.example                   # Environment variables template
└── README.md                      # This file

---

API Endpoints

Endpoint	Method	Description
/chat	POST	Main chat endpoint (hybrid: general + document Q&A)
/upload-pdf	POST	Upload and process PDF documents
/pdf-status	GET	Check if session has PDF loaded
/remove-pdf	POST	Remove PDF from current session

---

Development Commands

# Setup development environment
make setup

# Run Flask development server
make run

# Clean up (remove venv, cache, etc.)
make clean

# Install new Python dependencies
source venv/bin/activate && pip install package_name

# Frontend development
cd UI && npm run dev        # Development server
cd UI && npm run build      # Production build

---