Projects

A Python-based MCP server that gives AI tools safe, read-only access to fine-grained Supabase/Postgres schema metadata. It focuses purely on database structure: tables, columns, views, enums, RLS policies (and coverage), functions/RPC candidates, foreign keys, indexes, and triggers, so assistants can reason about migrations, audits, and relationship graphs without running arbitrary SQL.

A spaced repetition flashcard application built with Next.js. Study your flashcards with an intelligent algorithm that shows you difficult cards more often and easy cards less often. Features multiple decks, configurable settings, progress tracking, and CSV import functionality.

A web application for tracking university grades across different modules and calculating weighted averages. Features include semester and year breakdowns, module grade tracking, and comprehensive statistics.

Python package implementing gravitational n-body simulation with computational complexity optimisation. Developed direct O(n²) force calculation and Barnes-Hut algorithm implementation achieving O(n log n) complexity via quadtree spatial partitioning. Implemented custom quadtree data structure (Gadget) with recursive spatial decomposition for efficient force approximation. Designed modular package architecture with separate simulation classes, custom stack data structure, and physics-based body representation. Features animated visualisation of multi-body gravitational interactions with configurable time steps and astronomical unit scaling.

Browser extensions for LucidTrack enabling seamless job application tracking from job posting sites. Built Chrome extension utilising chrome.identity API for native OAuth authentication flow. Currently developing Safari Web Extension with tab-based OAuth redirect flow using Supabase callback handling due to Safari's lack of chrome.identity API support. Implemented manifest.json adaptation layer removing Chrome-specific permissions (identity, oauth2) for Safari compatibility whilst maintaining feature parity. Built automated build system with pnpm supporting both extension targets. Developed token extraction and authentication state management across browser contexts with automatic tab closure after successful OAuth completion.

Full-stack data cleansing application for normalising and deduplicating job application records. Built FastAPI backend with dual matching algorithms: RapidFuzz-based fuzzy string matching and sentence transformer embeddings for semantic similarity. Implemented composite embeddings matcher combining company name, position, and URL features for duplicate application detection. Architected entity assignment system linking companies to canonical entities with batch processing and transaction management. Developed URL cleansing pipeline normalising application links. Built Next.js 16 frontend with React Context state management and modular tab-based UI. Integrated Supabase for database operations with RPC functions for complex queries. Designed configurable threshold-based matching with real-time preview and export functionality.

EMS (Estate Management System) is an all-in-one platform centralising contact management, expense tracking, mileage and time logging, and estate inventory management. I made EMS to help my family manage their estate.

A silly 3D game where cube people run for their lives while you practice your aim. Built with Cursor.

A Python implementation of a Merkle Tree for learning and experimentation.

Designed and implemented a Java-based Horse Race Simulator. Featured multithreaded race execution for parallel processing and performance. Incorporated dynamic event handling to address real-time race conditions.

My personal portfolio website showcasing my projects and experience.

Real-time embedded firmware for ESP32 propulsion control utilising dual-core architecture with FreeRTOS task separation. Designed modular submodule architecture with interface-based abstraction (dependency inversion) to decouple sensor management from communication layers. Implemented non-blocking inter-core communication via FreeRTOS queues, preventing sensor data loss during high-frequency command processing. Built automated build system with Arduino CLI for dependency management and recursive submodule handling. Developed JSON-based command protocol with ESP ID validation and ACK messaging. Architected sensor abstraction layer supporting configurable polling intervals and extensible sensor implementations.

Asynchronus python application to coordinate communication between ESP32s, Dashboard, FSM (Finite State Machine). Non-blocking architechture with priortiy-based messaging from API calls.