Portfolio

Real problems, real solutions

Client systems I've replaced and personal projects I've built from scratch - each one a different kind of engineering challenge.

Screenshot placeholder - Carrycare dashboard

Enterprise System Replacement

Carrycare

A patient transport company with ~200 staff and ~90 vehicles, running on a split Access/SQL Server front-end, 102 Jotform forms, and 26 Power Automate flows. The existing system had a 233-column polymorphic customer table, free-text data entry errors, and cost thousands to add two fields to the crew portal.

I'm building a single web application that replaces the entire stack - every Access form, every Jotform submission, every automated flow - with a clean, modern system designed around how the business actually operates.

The Problem

  • 5 separate Access front-end databases
  • 102 Jotform forms with hardcoded dropdowns
  • 26 Power Automate flows for PDF filing
  • Crew portal charging thousands per change
  • Data scattered across 4 different systems

The Solution

  • 49 PostgreSQL migrations - clean schema
  • 35+ REST API routes (Fastify)
  • 30+ frontend pages (React/Mantine)
  • 10 built-in reports
  • Mobile-ready crew portal
TypeScript React Mantine Fastify PostgreSQL SQL Server Docker

Business Management System

W Coleman & Son

A family-run milk delivery company with 3,000+ customers, ~100 products, and 6 delivery rounds. They were running on MR8 - a legacy Delphi application using flat binary files - with no customer portal, no bank reconciliation, and no way for customers to manage their own orders.

I built a complete replacement: customer management, standing orders, weekly order generation, delivery tracking, invoicing with VAT and PDF output, bank CSV import with auto-matching, a four-tier admin system, and a full customer portal with magic-link authentication.

Screenshot placeholder - Coleman dashboard

The Problem

  • Legacy Delphi app with flat binary data files
  • No customer self-service
  • Manual bank reconciliation
  • Developer of original software controlling costs

The Solution

  • 22 PostgreSQL migrations, full data model
  • Electron desktop app with NSIS installer
  • Customer portal: signup, orders, holidays, invoices
  • Bank CSV auto-matching with configurable rules
TypeScript React Mantine Fastify PostgreSQL Electron Knex.js
Screenshot placeholder - Numbersmith homepage

Finance Calculator Website

Numbersmith

A UK finance calculator website with 60 calculators across 11 categories - mortgages, tax, savings, pensions, loans, and more. Zero dependencies, fast load times, proper schema markup, and SEO that competes with well-funded competitors.

The v2 redesign added dark mode, Chart.js visualisations, a shared CSS/JS architecture, copy/share/print toolbars, and full accessibility support.

60
Calculators
11
Categories
0
Dependencies
HTML/CSS/JS Cloudflare Pages Chart.js GA4 Schema.org SEO

Personal Projects

Built from scratch, for the challenge

Not every project has a client behind it. These are systems I built because I wanted to solve the problem myself - no frameworks, no shortcuts.

Screenshot placeholder - SCMail webmail

Self-Hosted Infrastructure

Electronic Mail

A full mail server and custom webmail client for my own domain. Postfix handles SMTP, Dovecot handles IMAP, OpenDKIM signs outbound mail, and a Cloudflare Tunnel bridges inbound delivery around my ISP's port 25 block.

The webmail frontend is a Python/Flask application with search, HTML rendering, attachments, drafts, signatures, keyboard shortcuts, and a PWA manifest for mobile. I also built an Android APK wrapper so it works as a native app on my phone.

What It Does

  • Full send/receive email on a custom domain
  • DKIM, SPF, DMARC - passes all deliverability checks
  • Cloudflare Worker + Tunnel for inbound relay
  • VPN split tunnelling for mail port routing

Webmail Features

  • Search, HTML rendering, attachments, drafts
  • Reply, reply-all, forward, star, bulk operations
  • Keyboard shortcuts, auto-refresh, signatures
  • PWA + Android APK wrapper (2.3 MB)
Python Flask Postfix Dovecot OpenDKIM Cloudflare Workers Let's Encrypt Android

Desktop Application

Mainframe

A personal database and full desktop environment built with Tauri v2 and React. All data is encrypted with SQLCipher (AES-256). It has its own file system, terminal emulator, code editor, image editor with layers, spreadsheet, budget tracker, calendar, contact manager, and project management - 180+ source files and 130+ backend commands.

The machine it runs on boots directly into Mainframe via a custom Ubuntu autoinstall ISO - LUKS full-disk encryption, no desktop environment, no browser, no escape. A locked-down single-purpose appliance.

Screenshot placeholder - Mainframe desktop
180+
Source Files
130+
Backend Commands
3-Step
Auth System
Tauri v2 Rust React TypeScript SQLCipher xterm.js LUKS Ubuntu Autoinstall
Screenshot placeholder - Lagnar editor

3D Game Engine

Early Days

Lagnar

A third-person RPG engine built from scratch in C++17 with SDL2 and OpenGL 3.3 - no Unity, no Unreal, no middleware. A 15km procedurally generated terrain with triplanar texturing, skeletal animation, shadow mapping, dynamic day/night and weather systems, water simulation, ragdoll physics, and an NPC schedule system.

The built-in level editor has 36 features including snap-point building, an interactive 3D gizmo, prefabs, layer management, structural integrity visualisation, terrain erosion and ramp tools, physics settle, and auto-save. 120+ procedural mesh assets generated by a companion tool.

Engine

  • OpenGL 3.3 renderer with 7 shader programs
  • Skeletal animation (24-bone, GPU skinning)
  • Shadow mapping, rim lighting, particle system
  • Ragdoll physics, water simulation, weather

Editor

  • 36 features: gizmo, prefabs, layers, snap-points
  • 9 editing modes inc. terrain sculpting and painting
  • 120+ procedural mesh assets (buildings, trees, rocks)
  • Multi-select, undo/redo, copy/paste, grouping
C++17 OpenGL 3.3 SDL2 GLSL Skeletal Animation Shadow Mapping Verlet Physics
Screenshot placeholder - HELLSEEKER gameplay

Game Engine & Editor

HELLSEEKER

A 2D action-platformer built in C and SDL2 - no Unity, no Godot, no engine. Custom physics with polygon collision meshes and slope handling, five enemy types with distinct AI behaviours, a boss with phase transitions, and an 8-projectile combat system with bombs, ropes, and thrown weapons.

The built-in level editor has five editing modes, undo history, copy/paste, a tile palette with Wang auto-tiling, entity placement, mechanism wiring, and a door-linked world system. Renders at 480x270 native resolution, upscaled 2x.

Engine

  • Polygon collision mesh with segment merging
  • 5 enemy types + boss with phase transitions
  • 8 projectile types, combat with knockback and i-frames
  • Mechanism system: platforms, gates, crushers, elevators

Editor

  • 5 editing modes: tile, entity, room, door, mechanism
  • Wang auto-tiling with per-cell toggle
  • 242 tile types, undo history, copy/paste regions
  • Door-linked world system with level manifest
C SDL2 Custom Physics Polygon Collision Level Editor AI

Automation & Data Pipeline

Job Hunter

A semi-automated job application pipeline. Scrapes listings from five sources (Reed API, Adzuna API, CV-Library, JobServe, Indeed), scores each one against a configurable keyword profile, generates tailored cover letters from six role-specific templates, and auto-applies via Playwright browser automation.

The web dashboard gives a filterable, sortable view of every scraped job with score, salary, and work type. Click a listing to see the full description, generated cover letter, and action buttons. So far it has scraped 2,300+ jobs, generated 1,700+ cover letters, and auto-applied to 600.

Screenshot placeholder - Job Hunter dashboard
2,300+
Jobs Scraped
1,700+
Cover Letters
5
Data Sources
Python Flask SQLite Playwright BeautifulSoup REST APIs Web Scraping
Screenshot placeholder - Mordecai

Computer Vision & Game AI

Early Days

Mordecai

A chess engine designed to play against a human on a real, physical board. A webcam watches the board, OpenCV detects the position after each move, and an alpha-beta search with a 627-position opening book decides the computer's response.

The end goal: a robot arm that physically picks up and moves pieces on the board, making the AI a tangible opponent sitting across the table. Fully self-contained - no cloud APIs, no subscriptions, just hardware and code.

Python OpenCV Alpha-Beta Search Computer Vision Robotics

Your project could be next

Whether it's a legacy system that needs replacing or a new idea that needs building, I'd like to hear about it.

Start a conversation