A collection of everything I've built — from production systems to university coursework to learning experiments.
End-to-end systems built for real-world use
Production-grade fraud detection with CatBoost, Feast, Kafka, and automated MLOps — sub-15ms latency.
Hybrid RAG system with Docling extraction, BM25 + vector search, cross-encoder re-ranking, and RAGAS evaluation.
Enterprise spam detection with DistilBERT, FastAPI, Streamlit, deployed on AWS SageMaker with Terraform IaC.
Automated serverless data pipeline ingesting Finnhub stock data via EventBridge and Lambda, transformed with Glue and Athena, visualised with QuickSight dashboards.
End-to-end recommendation engine with content-based filtering (TF-IDF), collaborative filtering (SVD), and popularity-based ranking on MovieLens 100K. Interactive Streamlit app with RMSE, Precision@10, NDCG evaluation.
Academic coursework and research projects
Full-stack campus operations system for facilities, bookings, maintenance tickets, and notifications with JWT + OAuth 2.0 authentication.
Carbon credit marketplace connecting project developers, verifiers, and buyers. Built the educational hub with structured learning modules and community resources.
Full-stack wildlife safari management system with CRUD operations, built as a team project.
Native Android application built with Kotlin as a solo university project.
Statistical analysis of Look-Ahead Range impact on traffic flow stability — hypothesis testing, regression, and visualisation in R.
Practical projects built to learn and experiment
End-to-end ML pipeline with experiment tracking, hyperparameter tuning, and reproducible pipelines.
Regression models for predicting forest fire severity using linear, ridge, and elastic net regularisation.
Principal Component Analysis for dimensionality reduction and visualisation on the Wisconsin breast cancer dataset.
Binary classification with logistic regression on a custom dataset, covering decision boundaries and evaluation metrics.
Regression model comparison — Random Forest, Linear, Ridge, Lasso, and Decision Tree. Random Forest performed best.
Hands-on practice with XGBoost for both classification and regression tasks.
Personal experiments and passion projects
This very site — a minimalist developer portfolio with AI assistant, dark/light theme, onboarding, and interactive design elements.
Formula 1 race winner prediction using XGBoost regression on historical race data.
Self-hosted homelab on an old laptop — Nextcloud, Jellyfin, monitoring with Prometheus and Grafana. Learned Linux, Docker, and networking.