DEEPGUARD AI
AI-generated image detection system
AI-generated image detection system built with EfficientNet, achieving 96.2% accuracy and 95.8% F1-score across modern generative image architectures. Improved generalisation through augmentation and regularisation, with a FastAPI inference service and Grad-CAM explainability.
- PYTHON
- PYTORCH
- EFFICIENTNET
- FASTAPI
- DOCKER
View Project on GitHubBUGINSIGHT
Issue severity & resolution prediction pipeline
Dual-task ML pipeline predicting GitHub issue severity and resolution time, achieving a 91.3% F1-score and 1.8-day RMSE. Improved prediction performance by 12% through NLP-based feature engineering and model optimisation, with FastAPI and automated GitHub API integration.
- SCIKIT-LEARN
- XGBOOST
- LIGHTGBM
- NLP
- FASTAPI
View Project on GitHubSMARTDIET AI
Nutrition recommendation & calorie prediction
Nutrition recommendation and calorie prediction system trained on 500+ dietary records, achieving 78% prediction accuracy. Reduced calorie prediction error by 15% through feature engineering and regression optimisation, integrated into a full-stack React and Flask application.
- PYTHON
- REACT.JS
- FLASK
- XGBOOST
- MONGODB
View Project on GitHubOIL WELL CHOKE
Autonomous choke control — Honeywell AI Hackathon
Autonomous oil well choke control system built for the Honeywell Industrial AI Hackathon using a data-driven Digital Twin and Model Predictive Control. Developed a predictive well model and closed-loop MPC controller to track production targets while maintaining pressure and operational constraints.
- PYTHON
- PANDAS
- SCIKIT-LEARN
- MPC
- DIGITAL-TWIN
View Project on GitHub