Portfolio

A command-line rhyme assistant built in Node.js for CM1035 Algorithms and Data Structures 1 midterm (University of London). Reads a word from the terminal and returns rhyme candidates using two complementary algorithms plus a merged view. Features: 📖 Suffix matching against a plain-text word list using custom stack and queue data structures 🔊 Phonetic matching against the CMU Pronouncing Dictionary (ARPAbet), comparing the last three phonemes of each entry 🔀 Merge step interleaves the two sorted candidate lists into a single sorted array ⚙️ Built on Node `readline` + `fs` only — no external dependencies 🧠 Vowel- and consonant-final suffix extraction via tokenised character streams Going beyond the spec: - Two independent matching algorithms (orthographic + phonetic) with a merged view, not just one - Hand-rolled stack and queue implementations rather than built-in arrays - CMUdict header-skipping logic that respects the standard first-entry marker

Website and online booking system for Batam 1 Wellness and Massage, the largest spa in Batam, Indonesia. Guests can browse treatments and facilities, build a booking cart, and reserve time slots online, with WhatsApp as a fallback channel. Features: 🛒 Booking cart: add up to 4 services per booking, pick duration, date, and check-in time ✅ Live availability check: every cart item is re-verified against the spa's booking system before confirming 📧 Booking confirmation page with email/SMS confirmation 🌏 Trilingual site: English, Chinese, and Indonesian 💬 Graceful WhatsApp fallback when online booking or the services menu is paused 🏛️ Facilities, FAQ, and contact pages with SEO meta tags per route

NutritionLM is an AI-powered nutrition companion designed to make healthy eating simple, personalised, and intelligent. Through conversational AI and smart tracking, it helps users understand, monitor, and optimise their nutrition effortlessly. Features: 🤖 AI Nutritionist Chat: Interact with an intelligent AI powered by Google Gemini 2.0 to get personalised nutrition advice, verify dietary information, compare perspectives, and analyse food images with context from your own uploaded documents. 📸 Smart Food Logging: Capture meals with a photo and let AI automatically detect ingredients, calculate detailed nutritional values, and assign a healthiness score, all stored in a searchable history. 📊 Analytics & Insights: Visualise your nutrition trends, track weekly progress, compare intake against goals, and stay motivated with streaks and achievements. 📚 Personal Knowledge Base: Upload your own dietary guidelines and documents, enabling AI to deliver personalised, source-backed answers using retrieval-augmented generation (RAG). 🔗 Seamless Integrations: Connect with Telegram for reminders and check-ins, and optionally sync with Google Fit for holistic health tracking. 👤 Personalised Experience: Set goals, dietary preferences, and restrictions through onboarding, and customise daily nutritional targets. Achievement: 1st Place.

Squiggle is a modern web app that transforms everyday chatting into something more meaningful — by making gift giving the heart of your conversations. Features: 🤖 AI for Recommendations: AI reads your chats and gets what you (or your friend) actually want. 🎁 Wishlist Management: Add your wishlist to your profile and friends team up to gift you exactly what you've been eyeing. ⚡️ Quick Join: Drop your budget, and we'll match you to the perfect group gift. Achievement: 2nd Place.

VeCook is an enhanced cooking companion built for Kitchen CoPilot. Perfect for anyone seeking guidance from an AI that cares about your health, corrects your mistakes, and understands you. Features: 🍽 Personalized Meal Planning: Powered by the Google Fit API, the AI dynamically adjusts recipes and ingredient portions based on your activity level for optimal nutrition. 🙌 Hands-Free Recipe Book: Navigate through cooking steps effortlessly using hand gestures — point right 👉 to move forward and left 👈 to go back (no more touching with messy hands). 🤖 AI-Powered Assistance: Made a mistake? Just raise your hand ✋ and ask the AI for guidance on how to fix it in real-time. 🎥 Record your cooking process and share it on social media!

MaiPanic is an all-in-one, serverless, and scalable crisis management superapp designed to handle situations ranging from personal emergencies to large-scale incidents. Whether it's a heart attack or a national emergency, MaiPanic provides users with the tools and information they need in critical moments. Features: 🤖 AI-Powered Guidance: Smart assistance to guide users during emergencies with relevant actions and information. 🎥 Emergency Tutorials: A library of quick-access videos to help users respond effectively in critical situations. 🆘 SOS System: One-swipe emergency trigger and incident reporting for immediate help. 👥 Family Status Tracking: Monitor the safety status of loved ones in real time. 📍 Live Incident Map: Stay updated with real-time crisis and incident feeds around you. 👤 Personal Profile: Store essential medical information and emergency settings for quick access. Achievement: Finalist (Top 14), CODE EXP Novice Track at BrainHack by DSTA.

A real-time computer-vision playground built in p5.js + ml5.js for the Graphics Programming Finals (University of London). The app processes live webcam input with multiple filters and an interactive pattern-lock game driven by hand-gesture detection. Features: 📸 Webcam capture with snapshot, 160×120 min resolution pipeline 🎨 Image processing: grayscale + brightness boost, RGB channel split with threshold sliders, CMY conversion, YCbCr conversion for luminance/chrominance separation 🙂 Live face detection with bounding boxes and per-face filters (grayscale, blur, CMY, pixelation) via keyboard shortcuts ✋ Hand-gesture pattern-lock mini-game: users trace heart, ribbon, and zigzag patterns with their index finger 🎚️ Performance-optimised blur (full-image blur, then face extract) to avoid harsh edges Going beyond the spec: - Pattern-lock extension combining memory and gesture control — usable as a group icebreaker - Custom YCbCr colour targeting tuned for better blue/red tone separation - Face-region-only filtering except for blur, which required a different approach to avoid harsh edges

OtoDecks is a dual-deck DJ desktop app built in C++ with the JUCE framework for the OOP Finals (University of London). Two independent players share a mixer, with waveforms, a playlist table, and a custom vinyl-jog UI. Features: 🎛️ Two independent decks sharing a `MixerAudioSource` ▶️ Per-deck controls: play/pause, loop, reset, volume, speed, low-pass and high-pass `IIRFilterAudioSource` 💿 Custom `VinylComponent` with rotation tied to playback position and drag-to-seek 🌊 `AudioThumbnail` waveform displays with click-and-drag seeking 📋 `TableListBox` playlist with multi-format import (WAV, MP3, FLAC), save/load, per-row load-to-deck, remove, and search filtering ⌨️ Spacebar toggles play/pause on both decks when main window is focused 🎨 `CustomLookAndFeel` subclassing `LookAndFeel_V4` for metallic vertical sliders and circular buttons Going beyond the spec: - Full pluggable audio graph: file decode → transport → resampling → low-pass → high-pass - Custom vinyl-jog control using `std::function` callbacks to keep UI decoupled from audio - Timer-based UI sync so waveform playheads and vinyl rotation track playback without tight coupling - Dual-build setup (Projucer + CMake) with background texture and bespoke styling

A full-stack event-management web app built for CM2040 Database Networks and the Web (University of London). Organisers draft and publish events with categorised tickets; attendees browse, filter, search, and book. Features: 🔐 Organiser auth with bcrypt + `express-validator` and session management 📝 Draft / publish workflow for events with categories and tags 🎟️ Full and concession ticket types with per-type quotas 📊 Organiser dashboard: bookings summary + remaining ticket counts 🔍 Attendee discoverability: category dropdown + keyword search (title or tag), combinable, filtered in SQL with bound parameters 🗓️ Timezone helpers, form validation, copy-link, and action confirmations in client JS Going beyond the spec: - Many-to-many tagging via `tags` + `event_tags` tables (ON DELETE CASCADE on both FKs) - Dynamic SQL builder combining `LIKE` title search, tag subqueries, and category filter in one statement - Organiser-side remaining-ticket display (not in the base spec) - Low-tech-debt structure: shared queries in `util/dbHelpers.js` and auth guards in `middleware/requireLogin.js`

VeSteps is a web app game that encourages users to walk. User's step count from Google Fit is converted into in-game currency where users can complete side quests, earn rewards and skins. Achievement: 2nd Place.

VeCycle is a platform that incentivises recycling. Users perform recycling actions, helping companies reduce their carbon footprint. In return, they earn B3TR tokens, and companies receive verifiable and transparent carbon offset data — essentially Recycle-to-Earn. Built on the VeChain Thor Blockchain for VeBetterDAO, integrating the OpenAI API for recycling action validation and React for the front end. Achievement: Placed in the top 11 teams out of 250 participants, advancing to the pitching session.

DSTA BrainHack 2024: TIL-AI Made an ASR (Automatic Speech Recognition) model using Wav2Vec2 and Whisper pretrained models. Utilised Vertex AI for model training and collaboration. Competed in the TIL-AI (Today I Learned AI) Novice Track. Achievement: Qualified to semi-final (Top 32 teams).

The objective was to ideate and build an AI model by integrating various Replicate.com AI models. The team built FengShuiGPT to advise users on how to re-arrange their furniture according to Feng Shui principles. Users take a picture of their room and FengShuiGPT provides text tips and suggestions, together with an AI-generated image for visualisation. Models used: meta/llama-2-70b-chat, yorickvp/llava-13b, stability-ai/sdxl.

Space Explorer is a side-scrolling platform adventure game built in p5.js for the ITP1 Final Coursework (University of London). You pilot a jetpack-wielding astronaut across platforms, collecting gems, dodging falling rockets, and racing to the flag pole. Features: 🎮 Side-scrolling platformer with smooth camera follow, lives system (3 hearts), scoring, and win/lose states 💎 16 collectable gems, 3 health kits, 5 canyons, and 6 falling rockets placed across the level 🚀 Animated jetpack character with distinct states (stand, walk, jump, fall) 🌌 Starry night scene with 2000 animated stars, drifting clouds, snow-capped mountains, and lava canyons 🔊 Full audio: looped ambient music, jump/fall/collect/drink/win/lose effects Going beyond the spec: - Factory pattern for platforms (9 across the map) with collision-based jumping - Constructor-based `RocketFall` enemies with per-instance speed and size - Particle-system victory sparklers using a custom `Sparkler` constructor - Preloaded, volume-optimised sound pipeline

A physics-based snooker web game built with p5.js and Matter.js for the Graphics Programming Midterm (University of London). Realistic ball physics, interactive cue stick, and four game modes including a unique wrap-around twist. Features: 🎱 Realistic ball-to-ball physics via Matter.js (zero-gravity table, momentum transfer, cushion/pocket collisions) 🎯 Four game modes: Normal, Random Reds, Full Random, and Table Wrap (balls teleport to opposite cushion) 🖱️ Mouse-aimed cue stick with up/down arrow power control and Enter-to-shoot 🟢 D-line cue ball placement with colour-coded validity feedback 🏆 Auto-scoring: reds removed on pocket, colours return to spot 📋 Real-time collision-type logging (cue-red, cue-color, cue-cushion) Going beyond the spec: - Table Wrap Mode — balls crossing a cushion teleport to the opposite side, impossible on a real table - No-overlap random placement algorithm for the free-placement modes - Clean OOP split across `Table`, `Ball`, `CueStick` classes with composed physics and rendering

A C++ OOP Midterm project (University of London) that analyses 40 years (1980–2019) of temperature data across 28 European countries and renders it as ASCII candlestick charts — a technique borrowed from financial markets. Features: 🌡️ CSV parsing for daily temperatures across 28 European countries 📊 Monthly and yearly OHLC (Open/High/Low/Close) candlesticks rendered as ASCII 🔍 Filter by country (trend over time) or by date (compare across countries) 📈 Bull/bear/doji trend markers for warming, cooling, and flat periods 🔮 Temperature prediction using Annual Mean Growth Rate (AMGR) 💻 Interactive menu-driven CLI Going beyond the spec: - Candlestick metaphor applied to temperature data, complete with trend classification - AMGR-based prediction algorithm projecting future OHLC values from historical data - Clean OOP separation across `WeatherEntry`, `Candlestick`, `ProcessData`, and `WeatherMain`

A Programming with Data coursework (University of London) exploring Indonesia's inflation in a global context and deriving a benchmark ROI for Indonesian investors who must beat both domestic inflation and IDR devaluation against major currencies. Features: 📈 Historical analysis 2000–2023 across USD, EUR, GBP, JPY, and CAD vs IDR 💹 Inflation data sourced from the World Bank; exchange rates from Investing.com / Yahoo Finance 🧮 CAGR-based calculations and average-rate comparisons for accuracy 🌐 Global-market index combining all five currencies for a holistic picture 🚫 Excludes the 1998 political instability period to avoid skewed baselines 📊 matplotlib / seaborn / plotly visualisations with anomaly highlighting 🖥️ Deployed as a notebook-style web page on Vercel Going beyond the spec: - Benchmark ROI target that accounts for both inflation and currency devaluation simultaneously - Multi-method verification (CAGR + average rates) to cross-check conclusions - Full web deployment so non-technical readers can explore the analysis

KantinPU is a website built to overcome the problem where school stalls didn't show their menu prices and pictures, making students unsure to try them. Led and initiated the whole competition — arranged general, stand-up meetings and consultations with teachers. Planned the overall website branding, UI/UX and features, made the website mobile-compatible, utilised GitHub for collaboration, designed 8 banners, was in charge of collecting product prices and photography, and directed the video presentation including scripting, video-taking, and editing. Competition held by Yayasan Sagasitas Indonesia with AWS at Provincial Level. Presentation Video: https://youtu.be/n9LlUWLbSfk Achievement: 1st Place, Best Video, Best Presentation.