home

Hi, I'm Kev! I plan software.

PROJECTS

Edo.

01/

CAAT

02/

Volleytube

03/

Transport Visualiser

04/

OTHER PROJECTS

05/

Pushy

Git rebuilt from scratch as a terminal program in pure POSIX shell: add, commit, branch, merge.

Shell

06/

Presto

A slide-deck builder in the browser: create, edit and present without leaving the tab.

JavaScript

07/

Pokémon DB

A relational Pokémon database, modelled and queried in raw PL/pgSQL.

PostgreSQL

08/

Sed

The core of the sed command reimplemented in Python: substitute, delete, print, address ranges.

Python

ABOUT

I'm a full-stack dev shipping ai native solutions to real problems at scale. Currently building purpl, from Sydney.

GET IN TOUCH

say hello → don't be shy ↗︎

Edo.

Edo is an app I'm currently building to immediately find the food choices and restaurants within my radius that best fit my fitness goals, whether I'm cutting, bulking, or just trying to be more health-conscious. Around that it handles the rest of eating out: full macro breakdowns for every dish, results ranked by how well they fit your targets, and an order builder that lands exactly on your numbers. Coming to the App Store soon.

Anyone serious about their fitness journey has to dial in their nutrition, and too often that means turning down meals out with friends, trading relationships for results. Edo is the app I envisioned to tackle both.

Under the hood it mixes verified nutritional databases with, for everything unverified, a 9-step pipeline that carefully estimates each dish's nutritional profile against AUSNUT, Australia's national food database.

features

Macro Radius: dishes ranked by fit across every restaurant in range.
Two-tier data: verified chain nutrition; transparent AI estimates elsewhere.
Dish detail: full breakdown with source attribution on every card.
Chain restaurant calculator: build an order from verified data only, with support for every Australian chain coming soon.

learnings

You never really build the data ingestion layer once and call it done. Every new menu, chain, and edge case sends you back to keep iterating on how dishes get parsed, matched, and estimated.

Keeping that layer honest means tightening the AUSNUT matching, catching bad estimates, and tuning each of the nine steps as real data exposes their gaps.

Tech Stack

React Native
FastAPI
Postgres
Redis

01 / macro radius

CAAT live demo →

CAAT, the College Application Assistance Tool, is a web portal that puts a student's entire university application in one place: schools, majors and scholarships matched to your profile, an application tracker with every deadline on one calendar, and built-in essay and resume editors so the writing lives next to the plan.

Applying for scholarships means living in ten different apps at once: every provider runs its own portal with its own account, forms, requirements and deadlines, and the schools themselves sit on yet more platforms. CAAT collapses all of that into a single place, with one profile, one tracker and one calendar, so nothing slips because it lived in the wrong tab.

Tech Stack

Next.js
TypeScript
Supabase
Tailwind

features

Application Tracker: every application, deadline and requirement on one dashboard.
School Search: 10,000+ universities, compared side by side.
Essay Workshop: guided prompts, version history and built-in feedback.
Scholarship Finder: 1,000+ scholarships, filterable by eligibility, tuition amount and more.
Resume Builder: guided templates, exported the way admissions offices expect.
Document Vault: every transcript and letter in one place, encrypted and shareable by link.

learnings

The hardest part was the data. University and scholarship pages don't want to be scraped: some sit behind scrape protection, so simple requests came back empty and I had to find ways around it, such as driving a real browser with Playwright, before a page would even load.

Every site also has its own HTML structure, so details like eligibility rules can't be pulled with one selector. In practice each new university broke the script in a new way, and every break sent me back to iterate on the parsing rules until the scraper could survive the whole list.

02 / essays

02 / applications

Volleytube app store link →

VolleyTube is a community sports app for Sydney volleyball: live scores, match threads, highlight clips and player profiles across the PVL, SVL and YSVL leagues, available on the App Store. One React Native + Expo codebase ships to iOS, Android and web on Supabase.

Volleyball NSW publishes results on clunky competition pages: no notifications, no history, no single place to check when you play next. The social life of the league happened in group chats and Instagram, completely disconnected from the actual results.

Tech Stack

React Native
Expo
Supabase
Postgres

features

Live results: scraped fixtures with set-by-set grids, pushed in realtime, plus manual score entry so attendees can fill scraper gaps.
Match threads: realtime comments with @mentions, replies and reactions on every game.
Video: highlight clips and YouTube recordings attached to each match, with official livestreams pinned.
Player identity: volleyball-specific profiles (position, height, spike reach) and a follow graph feeding a personalised home feed.
Share cards: any match captured as a 9:16 story card for the native share sheet.

learnings

One codebase for three platforms is a discount, not a freebie. Expo ships iOS, Android and web from the same React Native code, but the last stretch of share sheets, push notifications and safe areas still had to be handled per platform before it felt native anywhere.

The other lesson: push trust to the database. Row-level security on every writable table, crowdsourced scores funnelled through one validated RPC, and notifications and audit logs as Postgres triggers means every client stays correct and nothing can be bypassed in app code.

03 / my next match

Transport Visualiser live demo →

Transport Visualiser is an interactive route finder for the NSW train and metro network: click any two stations on a full-screen map and it plots the shortest path between them in real time, every line colour-coded from T1 through to the Metro, with pan and zoom across the whole network.

At uni you learn a concept, pass the exam, and by the semester after it's gone. This time I wanted to do something different: take something I'd actually learnt and apply it to the real world. So I took Dijkstra's shortest path algorithm from Data Structures and Algorithms (Shoutout COMP2521) and applied it across Sydney's railway network.

Under the hood I wrote my own graph class from scratch and converted the Transport NSW open dataset into the nodes and edges of a weighted directed graph, which Dijkstra's and A* then search for the shortest journey between any two stations.

features

Shortest path: routes computed in real time, with both Dijkstra's and A* supported.
Hand-rolled graph: a weighted directed graph class built from scratch, no library.
Interactive map: full-screen canvas, click-to-select stations, pan and zoom.
Line coding: every station coloured by its line, T1 through T9 plus Metro.

learnings

Third-party API datasets are not clean. The Transport NSW data arrived riddled with errors: typos in station names, stations duplicated under slightly different spellings, and each one silently corrupted the graph, splitting one real station into two disconnected nodes.

The lesson is to clean and process your data before you build on it. Deduplicating and normalising every station record first is what made the graph, and everything sitting on top of it, actually trustworthy.

Tech Stack

React
Konva
Flask
MongoDB

04 / network map

Hi, I'm Kev! I planbuild software.

Edo.

CAAT

Volleytube

Transport Visualiser

Pushy

Presto

Pokémon DB

Sed

Hi, I'm Kev! I plan software.