MCP Job Search System

An intelligent job search system using MCP (Model Context Protocol) with AI-powered job matching, supporting both English and Georgian languages. Features MongoDB data storage, PyTorch model training with LoRA fine-tuning, and comprehensive Docker infrastructure.

Chatbox

Database Monitoring

Quick Start

Prerequisites

• NVIDIA GPU with CUDA support (RTX 4050/4060 recommended)
• Docker & Docker Compose
• 8GB+ RAM
• 10GB+ free disk space

1. Clone and Setup

git clone https://github.com/your-username/usaqmuri-AI.git
cd usaqmuri-AI

2. Install System Dependencies (Arch Linux)

# Install CUDA toolkit
sudo pacman -S cuda cuda-tools

# Verify CUDA installation
nvidia-smi

3. One-Command Setup

cd devops
make install

This will:

• Build all Docker containers
• Start MongoDB with authentication
• Seed the database with sample data
• Launch the development environment

Monitoring Dashboards

After running make install, access your monitoring tools:

PyTorch Training Monitoring

make train        # Start training with monitoring

• TensorBoard: http://localhost:6006 (training metrics, loss curves)
• Jupyter Notebook: http://localhost:8888 (interactive experimentation)

Database Monitoring

• MongoDB Admin: http://localhost:8081 (username: admin, password: admin)
• Main App: http://localhost:8000 (job search interface)

Real-Time Monitoring

make logs         # View all container logs
make gpu-monitor  # Monitor GPU usage

GPU Optimization

RTX 4050 (6GB VRAM) - Optimized Configuration

Batch size: 1 with gradient accumulation (16)
Precision: BF16 for stability
Quantization: 4-bit for memory efficiency
Gradient Checkpointing: Enabled with reentrant compatibility
Estimated Training Time: ~93 minutes

Training Commands

# Start training with full monitoring
make train

# Quick test (development)
make train-test

# View training progress
make train-logs

MongoDB Integration

Database Structure

• Jobs Collection: Scraped job postings with full-text search
• Training Examples: AI training data with Georgian context
• User Interactions: Chat history and analytics
• Models: Model versions and performance metrics

Database Commands

make db-seed      # Populate with sample data
make db-backup    # Backup database
make db-restore   # Restore from backup
make db-stats     # View database statistics

Georgian Language Support

Specialized Features

• hr.ge and jobs.ge scrapers for Georgian job market
• Multilingual training data with Georgian context examples
• Location disambiguation (Georgia country vs. US state)
• Georgian company knowledge (TBC Bank, BOG, etc.)

Georgian Job Search Examples

"Find jobs in Tbilisi"
"მინდა ვიმუშაო საქართველოში" 
"IT jobs Georgia country"
"Georgian companies hiring developers"

Docker Infrastructure

Available Services

make dev          # Development environment
make prod         # Production deployment  
make train        # Training with monitoring
make scraper      # Automated job scraping

Service Overview

• app-dev: Development server with hot reloading
• mongodb: Database with authentication
• redis: Caching layer
• trainer: PyTorch training with GPU support
• mongo-express: Database admin interface
• nginx: Production reverse proxy

Development Workflow

Daily Development

# Start development environment
make dev

# View logs in real-time
make logs

# Run training experiments
make train

# Monitor GPU usage
watch -n 1 nvidia-smi

Model Training Workflow

# 1. Start training
make train

# 2. Monitor in browser
# - TensorBoard: http://localhost:6006
# - Jupyter: http://localhost:8888

# 3. View progress
make train-logs

# 4. Stop training
make train-stop

Performance Metrics

Training Performance (RTX 4050)

• Model: facebook/xglm-564M (564M parameters)
• Trainable: 2.35M parameters (0.41% via LoRA)
• Memory Usage: ~5.2GB VRAM
• Training Speed: ~1.2 samples/second
• Dataset: 118 examples (106 train + 12 validation)

System Requirements Met

PyTorch 2.6+ (Security compliance)
CUDA 12.4 (Latest driver support)
MongoDB 7.0 (High performance)
BF16 Precision (RTX 4050 optimized)

Model Training Details

Setup & Management

make install      # Complete setup
make dev          # Development mode
make prod         # Production deployment
make clean        # Clean containers/volumes

Training & Monitoring

make train        # Start training
make train-stop   # Stop training
make train-logs   # View training logs
make gpu-info     # Show GPU information

Database Operations

make db-seed      # Seed with sample data
make db-backup    # Backup database
make db-restore   # Restore database
make db-clean     # Clean database

Development Tools

make logs         # View all logs
make shell        # Access development container
make test         # Run test suite
make lint         # Code quality checks

Authentication Setup

# MongoDB credentials (development)
Username: admin
Password: devpassword123
Database: job_search

Contributing

1. Fork the repository
2. Create feature branch (git checkout -b feature/amazing-feature)
3. Commit changes (git commit -m 'Add amazing feature')
4. Push to branch (git push origin feature/amazing-feature)
5. Open Pull Request

License

MIT License - see LICENSE file for details.

---

** Ready to find your dream job with AI! Start with make install and explore the monitoring dashboards.**