Projects

A collection of AI/ML, software, and research projects I've built and shipped.

Loading…

ANPR System
✓ Completed
Computer Vision

Automatic Number Plate Recognition (ANPR) System

Multi-stage detection pipeline using YOLOv8 for vehicle identification and custom-trained model for precise licence plate localisation, achieving 95%+ accuracy. Integrated SORT algorithm for unique ID assignment and SciPy Linear Interpolation for occluded frame reconstruction.

Python YOLOv8 OpenCV SORT
Deep RL Game
✓ Completed
Game AI / Reinforcement Learning

Deep RL Agent for "The World's Hardest Game"

60FPS real-time browser-to-Python communication via WebSockets. DQN agent with Experience Replay across a 48-dimensional state space. Novel Euclidean reward function reduced average training time by 35% and improved stability across diverse game levels.

Python PyTorch JavaScript WebSockets
Multimodal VLM
✓ Completed
ML / LLM Fine-Tuning

Multimodal VLM Fine-Tuning Pipeline (Inovako)

Spearheaded fine-tuning of Llama 3.2 Vision and Qwen 2.5-VL to classify industrial materials from metadata and images with high precision. Implemented LoRA and 16-bit quantization to optimize a 3B parameter VLM, reducing memory footprint by 40% for mid-range GPU training.

Python LoRA Llama 3.2 Qwen 2.5-VL
Credit Risk
✓ Completed
ML / Financial

Credit Default Prediction ML Pipeline

End-to-end pipeline on 100K+ loan applications (Python + SQL). AUC of 0.88. Grid-search optimization of Random Forest and Gradient Boosting achieving 82% accuracy. Improved early-warning detection by 25%, leading to 15% reduction in bad-debt provision at BNP Paribas TEB.

Python SQL Scikit-learn Pandas
Recommendation System
✓ Completed
API / AI System

Multimodal Recommendation & Moderation System (Wibto)

Designed multimodal recommendation system that increased user retention. Automated detection for 98% of sensitive content and spam via AI-driven pipelines. Deployed with FastAPI backend and integrated into a cross-platform React + Swift ecosystem.

Python FastAPI TypeScript Swift
Portfolio Website
✓ Completed
Web

AI/ML Portfolio Website

This multi-page portfolio built with vanilla HTML, CSS, and JavaScript. Features particle animations, typewriter effects, glassmorphism cards, project search/filtering, EmailJS contact form, and a fully dark AI/ML design aesthetic.

HTML CSS JavaScript