$ ls ~/projects/

All Projects

A collection of security research, AI experiments, and infrastructure builds. Each project represents a deep dive into real-world problems.

crop_agent.py
Python TensorFlow React

Crop Disease Detection AI

v1.0.4 Production
def detect_disease(image_path):
    # Load pre-trained CNN model
    model = load_model("crop_disease_v1.h5")
    predictions = model.predict(preprocess(image_path))
    return classify_threat_level(predictions)

A multi-modal AI agent designed to help farmers identify crop diseases early through computer vision. The system processes uploaded images through a convolutional neural network trained on 50,000+ labeled agricultural samples.

Key Features

  • Real-time disease classification with 94% accuracy
  • Support for 12 major crop types (wheat, rice, corn, etc.)
  • Treatment recommendations based on detected pathogen
  • Progressive Web App with offline capabilities
  • Multi-language support (English, Hindi, Tamil)

Technical Stack

Backend

  • • Python 3.11 + FastAPI
  • • TensorFlow 2.x for inference
  • • Redis for caching
  • • PostgreSQL for user data

Frontend

  • • React 18 + TypeScript
  • • Tailwind CSS
  • • PWA with service workers
  • • Mobile-first design

Impact

Deployed to 500+ farmers in rural India during pilot phase. Reduced crop loss by an estimated 23% through early disease detection. Currently being evaluated for government agricultural extension programs.

scan_url.sh
Python Scikit-learn Shell

Phishing URL Detection Engine

v2.1.0 Active
./scan_url.sh --deep https://suspicious-site.com
> Analyzing URL patterns...
> Checking SSL certificate...
> Running ML classification...
> [THREAT DETECTED] Phishing probability: 97.3%

Machine learning-based URL analysis tool capable of identifying malicious websites in real-time. Built for my Introduction to AI/ML course, it uses feature extraction and ensemble learning to detect phishing attempts with high precision.

Detection Methods

  • URL structure analysis (length, special characters, TLD)
  • Domain age and registration verification
  • SSL certificate validation and chain of trust
  • HTML content similarity matching
  • Behavioral analysis using Random Forest classifier
  • Integration with VirusTotal and URLhaus APIs

Model Performance

96.2%
Accuracy
94.8%
Precision
0.02s
Avg Response

Dataset & Training

Trained on 150,000 URLs (75k legitimate, 75k phishing) sourced from PhishTank, OpenPhish, and Alexa Top 1M. Used stratified k-fold cross-validation and hyperparameter tuning via GridSearchCV.

root@hostinger
Docker Linux Nginx

Home Lab & VPS Infrastructure

Online • 42 days uptime
# System Status
Server: Online
Uptime: 42 days, 3:14:52
Services: Docker (12 containers), Nginx, Postfix
Security: Hardened, Fail2ban active
Location: Hostinger EU Datacenter

A self-hosted infrastructure running on a Hostinger VPS, managing personal cloud services, email server, and development environments. Fully hardened Linux setup with automated backups and monitoring.

Running Services

Mail Server

  • • Postfix + Dovecot
  • • SPF, DKIM, DMARC configured
  • • Spam filtering via SpamAssassin
  • • Encrypted with Let's Encrypt

Web Services

  • • Nginx reverse proxy
  • • Portfolio hosting
  • • Docker registry (private)
  • • Automated SSL renewal

Monitoring

  • • Prometheus + Grafana
  • • Custom alerting scripts
  • • Log aggregation
  • • Performance metrics

Security

  • • Fail2ban for intrusion prevention
  • • UFW firewall configured
  • • SSH key-only authentication
  • • Daily automated backups

Infrastructure Specs

  • Ubuntu 22.04 LTS (hardened kernel)
  • 4 vCPU cores, 8GB RAM, 160GB NVMe SSD
  • Dockerized microservices architecture
  • Automated deployment via GitHub Actions
  • Offsite backups to Backblaze B2

What I Learned

This project taught me Linux system administration, network security hardening, and the operational challenges of running production services. Managing my own mail server gave deep insights into email protocols, spam prevention, and DNS configuration.

cd ~/