Projects
From AI diagnostic platforms to low-level embedded systems — things I've built that I'm proud of.
Full-stack AI diagnostic platform. Upload a chest X-ray and enter patient vitals — three ML pipelines run in parallel: a ResNet-18 CNN classifying 15 chest conditions, a Gradient Boosting risk engine, and a Keras MLP fusion model, all unified into a real-time diagnostic report via FastAPI + React 19.
Enterprise-grade predictive analytics platform modelled on SAP Business Data Cloud. Features a Spring Boot REST API with JPQL aggregation, PostgreSQL 16, React 18 + Vite frontend, OLS linear regression demand forecasting, and full test coverage across all layers.
Real-time ASL finger-spelling recognition. Sign a letter — the model predicts it and spells out words live from your webcam. 4-layer MLP with batch norm, 10-frame smoothing window, and a 370k-word dictionary for real-time word suggestions.
Production-grade real-time analytics platform with multi-tenant architecture, JWT auth, Redis caching, and Prometheus/Grafana observability. Batch ingestion of 1,000 events/request with RFC 9457 error standards and full CI/CD via GitHub Actions.
Real-time anti-spoofing authentication system combining facial recognition with liveness detection to block deepfakes and photo spoofing. Multi-layer security pipeline with challenge-response protocol.
AI-powered mood journaling app that analyzes emotion trends over time. Built with React Native, NLP sentiment analysis, and a Node.js backend with personalized insights dashboard.
ML pipeline for real-time credit card fraud detection using ensemble methods. Handles severely imbalanced datasets with SMOTE oversampling, achieving high precision-recall under class imbalance.
Drives 4 LEDs on an Arduino Mega 2560 (ATmega2560) using hardware timer interrupts and software PWM to produce SOS morse code, breathing/fade effects, and choreographed light patterns — all with 1µs precision and non-blocking event-driven architecture.
Three optimized low-level algorithms in pure AVR assembly: BCD arithmetic with carry correction using the "add 6" technique, XOR-based Hamming distance calculation (bit-level edit distance), and rightmost consecutive bit sequence clearing — all operating directly on registers.
Converts 2D images into pixel graphs where each pixel is a vertex connected to same-color neighbors. Applies DFS/BFS traversal to perform flood fill, automatic boundary/region outlining, and connected component detection across up to 2000×2000 pixels.
Full implementation of Sedgewick's LLRB tree — a simplified Red-Black BST variant that enforces left-leaning invariant via rotations and color flips. Supports insert, delete, search, and floor/ceiling operations with O(log n) guarantee.
Interactive visualizer implementing Graham Scan and Jarvis March algorithms to compute convex hulls of point sets. Step-by-step animation shows the algorithm's decision process in real time with configurable point count.
Interactive cybersecurity-themed escape room game built for teaching security concepts. Players solve crypto puzzles, SQL injection challenges, and network forensics riddles to "escape" — making security education engaging and hands-on.
Cellular automata simulation engine exploring Conway's Game of Life and custom rule sets. Real-time interactive canvas with configurable grid size, speed, and initial conditions — observing how complexity emerges from simple rules.
Data dashboard for tracking biogas production and distribution metrics built in partnership with FortisBC. Real-time pipeline monitoring, historical trend charts, and anomaly detection — helping operators identify efficiency gaps.