10 projects spanning embedded firmware, computer vision, AI/ML, IoT, and full-stack development.
Offline Android clinical-reference app for the Navy Medical Service — on-device Gemma 3 1B RAG chat for Medical Officers, deterministic protocol search for medics, zero connectivity required.
| Platform | Android — model and vector database entirely on-device |
|---|---|
| Model | Google Gemma 3 (1B parameters) via MediaPipe LLM inference |
| Base | Built on Google's MediaPipe RAG Android sample |
| Retrieval | Hierarchical parent-child chunking — 1000-char parents, five overlapping ~250-char children |
Top-down factory collision-prevention system — custom-trained YOLOv8n detects humans and robots through an IP camera, with bounding-box overlap logic raising collision warnings.
| Dataset | Custom top-down dataset — 1,612 person + 2,132 robot usable instances (2 robot types + human mannequin) |
|---|---|
| Model | YOLOv8n (nano) — custom-trained |
| Detection | Top-down IP camera view; bounding-box overlap triggers collision warning |
| Role | R&D Engineer Intern — A*STAR ARTC |
Bare-metal C sleep-environment automation on PIC18F4550 — dual-RTC timekeeping over I²C, two-sensor activation logic, and a five-stage escalating wake alarm.
| MCU | PIC18F4550 |
|---|---|
| Toolchain | MPLAB X IDE / XC8 C Compiler |
| Peripherals | DS1307 RTC (I²C), 20×4 LCD, 4×4 keypad, PIR sensor, LDR, USB aromatherapy diffuser, buzzer |
| Context | Singapore Polytechnic — MAPP module |
JavaFX desktop wellness tracker with Harris-Benedict-based daily targets, gamified quiz and badge rewards, and a SOLID-structured OOP design — later migrated from local MySQL to Huawei Cloud RDS for the Huawei Developer Competition APAC 2023.
| Platform | Desktop — JavaFX with FXML and CSS |
|---|---|
| Database | MySQL 8 — migrated to Huawei Cloud RDS with Elastic IP binding |
| Targets | Harris-Benedict equation for personalized calorie, exercise, and sleep goals |
| Design | MVC + SOLID — separated DB handler, calculator, and validator classes |
Android AR real-estate app with ARCore furniture placement during home tours, a loan simulator, and Maps integration.
| Platform | Android (Java) |
|---|---|
| AR Engine | ARCore — furniture placement during physical home tours |
| Features | Loan simulator, Maps API, SQLite listings |
| Design | Full UI/UX flow designed in Figma before development |
Keypad-driven FSM control program simulating a multi-phase wash cycle on a Raspberry Pi CM3 (ARM/Linux) — timer-driven state transitions, hatch-open safety interlock, and stepper-motor drum simulation.
| Platform | Raspberry Pi Compute Module 3 — ARM Cortex-A, embedded Linux |
|---|---|
| Toolchain | GCC ARM cross-compiler (arm-linux-gnueabihf), VS Code remote debug via gdbserver |
| Peripherals | 4×4 keypad, 20×4 LCD, 7-segment display, stepper motor, buzzer |
| Context | Singapore Polytechnic — Embedded Computer Systems module, 2023 |
C++ bin-packing cost-minimizer using exhaustive recursive permutation search — guarantees global minimum storage cost and wastage.
| Language | C++ |
|---|---|
| Algorithm | Recursive permutation search (exhaustive) |
| Input | Rack weight limits and per-package costs, read from a text file |
| Output | Global minimum total cost, racks used, and wastage |
Full-stack charity platform with PayPal sandbox donations, per-centre community forums, and Maps integration.
| Stack | PHP / MySQL / JavaScript |
|---|---|
| Payments | PayPal sandbox integration — personal-to-business transfer |
| Features | Per-centre forums, database-driven search, dynamic gallery |
| Team | 4 members |
Flask web app for self-directed phobia detection and exposure training using OpenCV — built in 5 days at MindfulHacks 2022, learning Python, OpenCV, and Flask from scratch.
| Stack | Python / OpenCV / Flask |
|---|---|
| Scope | Scopophobia trainer — architecture extensible to additional phobias |
| Timeline | 5 days — Python, OpenCV, and Flask learned from scratch |
| Context | MindfulHacks 2022 — first hackathon |
IoT accessibility concept — a wearable token with Wi-Fi, Bluetooth, GPS, and an SOS button that gives visually impaired commuters real-time audio guidance at bus stops and MRT stations.
| Concept | Wearable token + retrofitted signalers on bus stops, buses, and MRT platforms |
|---|---|
| Token hardware | Wi-Fi module, Bluetooth transmitter/receiver, GPS module, SOS button |
| Features | P2P audio announcements, bus-arrival alerts, SOS with GPS-located dispatch |
| Team | 3 members — Singapore Polytechnic + NUS |