🎭 AI-Powered Real-Time Affective Chatroom

A modern, intelligent chat platform that goes beyond text by understanding the emotional tone of conversations in real-time. This project uses a microservices architecture to detect emotions, moderate toxic content, and generate conversation summaries using advanced AI models.

---

🚀 Features

• 💬 Real-Time Messaging: Instant communication using WebSockets with zero latency.
• 🤩 Real-Time Emotion Detection: Messages are automatically annotated with emojis representing 27 distinct emotions (e.g., Joy, Admiration, Fear) using the modernbert-base-go-emotions model.
• 🛡️ AI-Powered Moderation: Automatically detects and blocks toxic content using roberta_toxicity_classifier, keeping the community safe.
• 📊 Smart Insights: Generates on-demand chat summaries and analyzes the overall "Room Mood" using Google Gemini 1.5 Flash.
• ⚡ Asynchronous Architecture: Uses a "Broadcast-First, Update-Later" pattern to ensure the chat feels instant even while AI processing happens in the background.

---

🏗️ System Architecture

The system is built as a set of decoupled microservices deployed on Google Cloud Run, connected to a Google Cloud SQL (MySQL) database.

Microservices

1. chat-app-service (FastAPI): The core orchestrator. Handles WebSockets, user authentication, database management, and calls other AI services.
2. ml-emotion-service (FastAPI + Transformers): A dedicated high-memory service that runs the Emotion Analysis model.
3. ml-toxicity-service (FastAPI + Transformers): A dedicated high-memory service that runs the Toxicity Detection model.

---

🛠️ Tech Stack

• Backend: Python 3.10, FastAPI, Uvicorn, Gunicorn

• Frontend: HTML5, JavaScript (ES6), Tailwind CSS

• Database: MySQL (Google Cloud SQL), SQLAlchemy (ORM)

• AI & ML:

  • Hugging Face Transformers (modernbert, roberta)
  • LangChain + Google Gemini (LLM)
  • PyTorch

• DevOps: Docker, Google Cloud Run, Google Artifact Registry

---

⚙️ Local Development Setup

Follow these steps to run the project locally.

Prerequisites

• Python 3.10+
• Docker (optional)
• MySQL Server (local or cloud)

1. Clone the Repository

git clone https://github.com/your-username/affective-chatroom.git
cd affective-chatroom

2. Set Up Virtual Environment

python -m venv chat_env
# Windows
./chat_env/Scripts/activate
# Mac/Linux
source chat_env/bin/activate

3. Install Dependencies

pip install -r chat-app/requirements.txt
pip install -r ml-emotion/requirements.txt
pip install -r ml-toxicity/requirements.txt

4. Configuration (.env)

Create a .env file in the chat-app/ directory:

DB_USER=root
DB_PASS=your_local_password
DB_HOST=127.0.0.1
DB_NAME=chatroom
SECRET_KEY=your_secret_key
GOOGLE_API_KEY=your_gemini_api_key

EMOTION_API_URL=http://127.0.0.1:8001
TOXICITY_API_URL=http://127.0.0.1:8002

5. Run the Services (3 Terminals)

Terminal 1 — Emotion Service

cd ml-emotion
python main.py  # Runs on port 8001

Terminal 2 — Toxicity Service

cd ml-toxicity
python main.py  # Runs on port 8002

Terminal 3 — Main Chat App

cd chat-app
uvicorn main:app --reload --port 8000

Access the app at: http://127.0.0.1:8000

---

☁️ Deployment (Google Cloud)

1. Deploy AI Services (High Memory)

gcloud run deploy ml-emotion-service --source ./ml-emotion --allow-unauthenticated --memory=4Gi --region=asia-south1
gcloud run deploy ml-toxicity-service --source ./ml-toxicity --allow-unauthenticated --memory=4Gi --region=asia-south1

2. Deploy Chat App (Standard Memory)

gcloud run deploy chat-app-service \
  --source ./chat-app \
  --allow-unauthenticated \
  --memory=1Gi \
  --region=asia-south1 \
  --set-env-vars="DB_USER=root,DB_PASS=YourDBPass,DB_NAME=chatroom,DB_HOST=YOUR_CLOUD_SQL_PUBLIC_IP,SECRET_KEY=secure-key,GOOGLE_API_KEY=your-key,TOXICITY_API_URL=https://url-to-toxicity-service,EMOTION_API_URL=https://url-to-emotion-service"

---

🔮 Future Scope

• Mobile App using React Native
• Voice emotion recognition
• End-to-end encrypted private chatrooms

---

📄 License

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