AI Chat Hub

AI-Powered Distributed Real-Time Chat Application

A full-stack, real-time chat application with JWT authentication, live WebSocket messaging, MySQL persistence, and AI-powered Smart Reply suggestions. Built as a two-project monorepo — a Spring Boot API/WebSocket backend and a static HTML/CSS/JS frontend — fully decoupled and independently deployable.

Backend/    Spring Boot 3 + WebSocket (STOMP) + Spring AI + MySQL (Java/Maven)
Frontend/   Static HTML/CSS/JS client (no build step, deploys as-is)

---

Table of contents

• Overview
• Problem statement
• Objectives
• Key features
• Technologies used
• System architecture
• Frontend architecture
• Backend architecture
• Database architecture
• JWT authentication
• WebSocket real-time communication
• AI / Smart Reply architecture
• Folder structure
• Installation requirements
• Local setup
• MySQL setup
• Environment variables
• Running the application
• API overview
• WebSocket overview
• Smart Reply workflow
• Production deployment
• Troubleshooting
• Security considerations
• Future enhancements
• Running tests

---

Overview

AI Chat Hub is a two-person-to-two-person (1:1) real-time messaging application: users register, log in, see who else is registered, and chat with them instantly over a WebSocket connection, with full message history persisted in MySQL. An AI layer (OpenAI or a local Ollama model, interchangeably) powers two features on top of that: Smart Reply (context-aware suggested responses inside a conversation) and a standalone AI Assistant chat page.

Problem statement

Most "build a chat app" tutorials stop at either the real-time messaging part or the AI part, rarely both, and rarely with production concerns (auth, CORS, deployment, secret handling) treated seriously. This project sets out to build a chat application that is simultaneously:

• Real-time — messages must arrive instantly, not on a polling delay.
• Persistent — a message survives a refresh or logging in from a different device/browser.
• Secure — every API and WebSocket connection is authenticated; no secret is ever hardcoded or exposed to the client.
• AI-augmented — without letting the AI layer become a single point of failure for the rest of the app.
• Actually deployable — configured from day one via environment variables rather than hardcoded localhost values that only work on one machine.

Objectives

1. Implement secure user registration and login backed by hashed passwords and stateless JWT authentication.
2. Implement real-time, bidirectional messaging over WebSocket (STOMP), authenticated the same way as the REST API.
3. Persist every message and make full conversation history retrievable, independent of any single browser's local storage.
4. Integrate an AI provider to generate short, contextual reply suggestions from real conversation history, with a provider that can be swapped (OpenAI ↔ local Ollama) without code changes.
5. Ensure AI failures degrade gracefully — normal chat must never break because the AI layer is slow, down, or unconfigured.
6. Keep the whole system configurable purely through environment variables, so the same codebase runs unmodified in local dev and in production.

Key features

• Authentication — register/login with bcrypt-hashed passwords, JWT issuance and validation, server-side logout (token revocation, not just a client-side redirect), profile editing (name + password change).
• Real-time messaging — WebSocket chat via STOMP/SockJS: instant delivery, typing indicators, online/offline presence (correct across multiple open tabs/devices), read receipts, per-conversation unread counts.
• Message persistence — every message is stored in MySQL and survives a refresh or a login from a different device (GET /api/chat/history, GET /api/chat/conversations).
• AI Smart Reply — three short, context-aware, tonally distinct reply suggestions generated from the last 10 messages of a conversation, with a graceful static fallback if the AI provider is slow, down, or not configured.
• AI Assistant — a standalone free-form chat page backed by the same configurable AI provider.
• Contacts, Chat History, Profile, Settings, About pages, all backed by real API calls — no mock data.

Technologies used

Layer	Technology
Backend	Java 21, Spring Boot 3.5, Spring Security, Spring Data JPA, Spring WebSocket (STOMP), Spring AI (OpenAI + Ollama), JJWT, BCrypt, Maven
Database	MySQL 8 (Hibernate ddl-auto=update)
Frontend	HTML5, CSS3, vanilla JavaScript (ES2017+), SockJS, @stomp/stompjs — no framework, no build step
Testing	JUnit 5, Spring Boot Test, AssertJ, H2 (in-memory, test-only)
Deployment	Vercel (frontend, static), any JVM host — Render/Railway/Fly.io/EC2 (backend)

This is the beginning of the README. Read the rest at the original. Read the full README ↗