🚀 LLaMA — Full-Stack AI Chat Application

A production-grade AI-powered chat application built using Next.js, MongoDB, OpenRouter LLMs, and secure authentication, featuring persistent chat history, PDF-based AI conversations, context-aware responses, and scalable architecture.

🔗 Live Demo: https://llama-theta.vercel.app/
💻 GitHub: https://github.com/Kartikey-Pathak/LL

---

📌 Overview

LLaMA is a full-stack AI chat platform where users can:

• Securely authenticate
• Create and manage multiple chat sessions
• Chat with an AI model using conversation history
• Upload PDFs and ask questions from documents
• Persist chat history across sessions

The project was built to gain hands-on experience with:

• Real-world authentication systems
• AI integration in production apps
• Database schema design
• Stateful chat architectures
• Cost-aware LLM usage
• Deployment & serverless limitations

---

✨ Features

💬 AI Chat System

• Context-aware AI responses
• Previous messages used as conversation memory
• Multiple chat sessions
• Automatic title handling
• Persistent chat history
• Chat deletion support

📄 PDF Chat (Document Q&A)

Users can upload PDFs and ask questions directly from them.

Current Workflow

• Upload PDF
• Extract text from document
• Store parsed PDF content in MongoDB
• AI answers questions using uploaded document context

Safeguards

• Daily PDF upload limit
• Duplicate PDF detection
• File size restrictions
• User-specific document storage

⚠️ Note: Since the app currently uses a free LLM model, responses based on PDFs may occasionally be inaccurate or hallucinated.

---

🔐 Authentication & Security

Supports two authentication systems:

Email + Password Auth

• Signup/Login
• OTP email verification
• Password hashing using bcrypt

OAuth Authentication

• Google Sign-In support via NextAuth

Security Features

• JWT authentication
• HTTP-only cookies
• Protected API routes
• User-specific data isolation
• Authentication middleware

---

🧠 AI Integration

Powered by OpenRouter API with:

• Context-aware responses
• Controlled message history for lower token cost
• Cost optimization using limited context window
• Streaming response support

Current Model

• meta-llama/llama-3-8b-instruct
• Knowledge cutoff depends on selected model

---

⚛️ Frontend & UX

• Modern responsive UI
• Light/Dark mode
• Smooth chat experience
• Animated input system
• Toast notifications
• Auto-scroll to latest message
• Loading states & upload feedback

---

🛠️ Tech Stack

Frontend

• Next.js (App Router)
• React
• Tailwind CSS
• DaisyUI
• Framer Motion / Motion
• React Hot Toast

Backend

• Next.js API Routes
• MongoDB
• Mongoose
• JWT Authentication
• NextAuth
• bcrypt

AI & Document Processing

• OpenRouter API
• LLaMA Models
• PDFReader for PDF parsing

---

🗂️ Backend Architecture

Structured with modular architecture for scalability.

API Routes

/app
 ├── api
 │   ├── auth
 │   ├── verify
 │   ├── chatai
 │   ├── upload-pdf
 │   └── auth/[...nextauth]

Challenges Solved During Development

Some real-world engineering problems solved while building:

• AI context handling
• Persistent chat memory
• Duplicate chat title handling
• PDF parsing & storage
• Authentication with JWT + OAuth
• Serverless deployment issues on Vercel
• File upload limits (413 Payload Too Large)
• Error-safe API response parsing
• Token cost optimization

Learning & Development

• This project was built as a deep learning experience in:

• Full-stack development

• Authentication systems

• AI integration

• Backend architecture

• Database modeling

• Production debugging

• Resources used:

• Next.js Documentation

• MongoDB Documentation

• Stack Overflow

• ChatGPT (for debugging, explanations, and learning concepts)

• The core implementation, debugging, architectural decisions, and understanding were done during the development process.

Author

Kartikey Pathak

B.Tech CSE | Full-Stack Developer | Coding Enthusiast