00 · Hero

01 · frunt · main project

03 · Other work

02 · MGKFitness

05 · About

04 · How I work

06 · Contact

Matthew Kay

Kay

357 × 446

2,420 contributions in the last year

frunt

The problem

Restaurants have their rules written down. Staff don't read them.

I saw it working behind a bar: training was a compliance module that ticked a box, and finding an answer meant flipping through spec sheets or asking a manager again.

What it does

A manager uploads the documents the restaurant already has.
frunt turns them into short courses on staff phones.
Staff ask in plain words; every answer names the document it came from.

For independent restaurants · built and run solo · live since June 2026

frunthospitality.com, the frunt website — Website · hover to scroll

frunt manager dashboard, home — Manager web app · Home

01Ask

Staff ask in plain words, in the app or on WhatsApp. Every answer lists the documents it came from, so a manager can check it.

02Training

Short courses built from the restaurant's own documents, not a generic module. Each one finished is recorded, with the date it's due again.

01Manager appLive

Documents in; cited answers and training out.

Next.js · Supabase pgvector · Claude · Inngest · Stripe

02Staff appLive

Ask, read, sign off and train from a phone, or ask on WhatsApp. On the App Store and Google Play.

Flutter · Firebase messaging

03WebsiteLive

Marketing site, pricing and ten UK-law guides, built to be found.

Next.js · structured data · llms.txt

04Admin consoleIn use

How I run the business: customers, revenue, AI cost per feature, health checks.

Stripe · PostHog · Sentry · GitHub + Vercel APIs

05Instagram studioLive

Claude proposes posts and I approve them; they render, schedule, publish and report back.

Instagram API · Inngest · Satori

06Outreach + analyticsIn use

Sourcing, email, calls and visits, with PostHog and Search Console showing what works.

Notion · Gmail · Vercel cron · Search Console

frunt staff appLive · actively updated

MGK

Fitness

Liftio 1.4 home screen — Liftio 1.4 · React Native

Liftio 1.4 exercise detail with a progress chart

Lift 2.0 logging a workout — Lift · Flutter

Decision 0001 · accepted 17 Aug 2026

Liftio is replaced, not relaunched.

"Retiring a live record to avoid an untidy string is paying a real cost for a cosmetic one."

LiftLive · actively updated

RunLive · actively updated

mgkfitness.mgkcodes.com, the MGKFitness website — Website

01RunLive

Running tracker with an AI training coach.

Flutter · GPS · shared design system

02LiftLive

Strength log with a coach of its own. The rebuild of Liftio.

Flutter · shared design system

03Backend + AI coachLive

One account across both apps. The coach plans around your whole week.

Supabase · Edge Functions · OpenRouter · RevenueCat

04WebsiteLive

Scroll film, waiting list, support and legal pages. No tracking, by design.

Next.js 16 · Replicate clips

05SocialIn development

Posts drawn in Remotion from the site's own words, published through the Instagram API.

Remotion · Instagram API

06Release opsIn use

Store listings and submissions run from Claude Code; builds on Codemagic.

App Store Connect · Google Play · Codemagic

Other

Work

Ledger finance dashboard — Ledger · web appBusiness finance dashboard with AI transaction entry.

MSA architecture portfolio — MSA · client sitePortfolio for an architecture student.

Red Cross Reigate pub site — Red Cross · client siteLanding page for a pub in Reigate.

YouTube clone home feed — YouTube clone · practiceFull-stack video platform with auth.

Netflix clone home — Netflix clone · practiceState management, layer by layer.

FootyScoresLive scores, React Native + Node.

FootyScores app, live Premier League scores — FootyScoresLive scores, React Native + Node.

About

Building things has always been how I think. Before I knew what programming was, I was pulling things apart to see how they worked.

I run MGKCodes, the studio behind frunt and MGKFitness.

MGKCodes · the studio
behind frunt + MGKFitness

How I

Work

01Plan

Every screen laid out on one board, each with what it's for.

e.g. Lift screen board, v11

02Decide

Big calls written down with the reasoning, so they can be checked.

e.g. Decision 0001

03Build

AI does the typing. My own skills library sets how it works.

e.g. github.com/MattKay02/skills

04Check

Screens checked by eye, design reviews, a build check on every PR, and most of my skills.

e.g. 19 findings, 30 Sept review

05Ship

Store submissions with a written record of what went out.

e.g. Run 1.0, 2 Oct 2026

Inthe running app
Doesevery screen, in reading order
Outone board that updates in place

Plan · Product MapSource screen-board

Lays every screen of an app on one board, each with what it's for.

Inreal screens, type and marks
Doesoptions rendered in Remotion
Outside by side on one page

Decide · Motion DesignSource motion-sketchbook

Sketches design options as films and stills, so the choice is made by watching.

Decision records

Every big call written down: what I chose, the obvious alternative, what it costs, and what would change my mind.

Open source8 Claude Code skills

How the AI works for me is written down as skills: procedures with the checks built in. Anyone can install them.

npx skills add MattKay02/skills

A real walk of the CV viewer on matthewkay.dev, 6 Oct 2026: six numbered screenshots (opened, scrolled to the end, dark theme, a phone opening the /#cv link first, the standalone page) and the one finding, a heavy focus ring on the phone close button.

Check · UI EvidenceSource ui-walk

Screenshots every state of a feature, in order, ready to flip through.

Ina page or screen
Doesseven axes, checked against the code
OutP1, P2 and P3, with file:line

Check · UI ReviewSource page-audit

Reviews a page on seven axes and returns a ranked punch list.

Ina Flutter change
Doesan adb screenshot
Outread back and checked

Check · Mobile QASource emulator-verify

Checks a Flutter change on the Android emulator by reading the screen back.

Lighthouse on matthewkay.dev, mobile, median of three runs (77, 78, 80), 6 Oct 2026: Performance 78, Accessibility, Best practices and SEO 100. Fix first: LCP 5.1 s, mostly the hero name waiting on its fade-in; defer 21 offscreen images (about 528 KiB); preconnect to the live skills data (about 330 ms).

Check · Web PerfSource lighthouse-audit

Measures a production build on mobile and says what to fix first.

The sweep on a test service with five planted silent failures, all five found. Top three of eight ranked findings: allergens show as none when the database fails; orders can get stuck, run twice or crash the queue; Download all orders stops at 1,000.

Check · ReliabilitySource silent-failure-sweep

Finds failures nobody is told about, ranked by who they hurt.

Ina branch
Doesa Codemagic build, confirmed first
Outthe result and download links

Ship · Mobile CISource codemagic-build

Starts a CI build from the terminal and reports back when it lands.

Let's

Talk

mattykay2002@gmail.com

MK.Matthew Kay
Product engineer

The tour, as text

Intro

Product engineer. I design and build polished apps, end to end.

frunt staff app: Live · actively updated, App Store, Google Play
Run: Live · actively updated, Google Play
Lift: Live · actively updated, Google Play

The whole board

Everything I've built, laid out the way I work on it.

Scroll for the guided tour, or click any frame to jump straight to it.

frunt: the problem

Restaurants have their rules written down. Staff don't read them.

frunt turns a restaurant's own documents into staff training and instant answers. A web app for managers; a phone app and WhatsApp for staff.

frunt: the manager app

Managers bring the documents they already have.

frunt reads them, files them and turns them into courses, and a manager checks each one before staff see it. Use the tabs under the laptop to look through the screens.

frunt: sourced answers

Every answer shows where it came from.

Staff ask in plain words. frunt answers only from that restaurant's documents and cites each source, so a manager can check it.

frunt: the staff app

Staff get answers and training on their phone.

Two screens from the staff app, on the App Store and Google Play. Staff can also ask on WhatsApp, with no app to install.

frunt: the whole system

More than an app.

frunt is six parts I designed, built and run: two apps, a website, an admin console, an Instagram studio and an outreach engine.

Manager app (Live): Documents in; cited answers and training out.
Staff app (Live): Ask, read, sign off and train from a phone, or ask on WhatsApp. On the App Store and Google Play.
Website (Live): Marketing site, pricing and ten UK-law guides, built to be found.
Admin console (In use): How I run the business: customers, revenue, AI cost per feature, health checks.
Instagram studio (Live): Claude proposes posts and I approve them; they render, schedule, publish and report back.
Outreach + analytics (In use): Sourcing, email, calls and visits, with PostHog and Search Console showing what works.

MGKFitness: Liftio

Liftio. My first App Store app.

A gym tracker in React Native: 330+ exercises, progress charts, personal bests and cloud backup.

MGKFitness: the rebuild

Then I rebuilt it as part of something bigger.

Lift is Liftio rewritten in Flutter, one app in a suite with Run. I kept the old App Store listing instead of starting fresh: its reviews and history were worth more than a tidy app ID.

MGKFitness: the suite

Two apps that know about each other.

Log a run in Run and Lift sees it, so the coach plans around your whole week.

MGKFitness: the whole system

A suite, not two apps.

Two apps on one account and one backend, an AI coach, a website, a social pipeline and the release tooling around them.

Run (Live): Running tracker with an AI training coach.
Lift (Live): Strength log with a coach of its own. The rebuild of Liftio.
Backend + AI coach (Live): One account across both apps. The coach plans around your whole week.
Website (Live): Scroll film, waiting list, support and legal pages. No tracking, by design.
Social (In development): Posts drawn in Remotion from the site's own words, published through the Instagram API.
Release ops (In use): Store listings and submissions run from Claude Code; builds on Codemagic.

Other work

Smaller builds, each made to learn something.

Client sites, experiments and practice projects. Hover a card to see it in colour.

How I work

I use AI to move fast. These are the checks that keep it honest.

Plan, decide, build, check, ship. Each step leaves something behind that you can look at, and under each one are the skills I built for it, loaded live from GitHub.

Install my skills: npx skills add MattKay02/skills (MattKay02/skills)

codemagic-build (Ship): Starts a CI build from the terminal and reports back when it lands. Tested · loads when asked and stays out otherwise in 2 of 2 cases.
ui-walk (Check): Screenshots every state of a feature, in order, ready to flip through. Tested · loads when asked and stays out otherwise in 2 of 2 cases. Shown: A real walk of the CV viewer on matthewkay.dev, 6 Oct 2026: six numbered screenshots (opened, scrolled to the end, dark theme, a phone opening the /#cv link first, the standalone page) and the one finding, a heavy focus ring on the phone close button.
page-audit (Check): Reviews a page on seven axes and returns a ranked punch list. Tested · loads when asked and stays out otherwise in 2 of 2 cases.
emulator-verify (Check): Checks a Flutter change on the Android emulator by reading the screen back. Tested · loads when asked and stays out otherwise in 2 of 2 cases.
lighthouse-audit (Check): Measures a production build on mobile and says what to fix first. Tested · loads when asked and stays out otherwise in 2 of 2 cases. Shown: Lighthouse on matthewkay.dev, mobile, median of three runs (77, 78, 80), 6 Oct 2026: Performance 78, Accessibility, Best practices and SEO 100. Fix first: LCP 5.1 s, mostly the hero name waiting on its fade-in; defer 21 offscreen images (about 528 KiB); preconnect to the live skills data (about 330 ms).
screen-board (Plan): Lays every screen of an app on one board, each with what it's for. Tested · loads when asked and stays out otherwise in 2 of 2 cases.
silent-failure-sweep (Check): Finds failures nobody is told about, ranked by who they hurt. Tested · loads when asked and stays out otherwise in 2 of 2 cases · task score 0.92 with the skill, 0.88 without. Shown: The sweep on a test service with five planted silent failures, all five found. Top three of eight ranked findings: allergens show as none when the database fails; orders can get stuck, run twice or crash the queue; Download all orders stops at 1,000.
motion-sketchbook (Decide): Sketches design options as films and stills, so the choice is made by watching. Tested · loads when asked and stays out otherwise in 2 of 2 cases.

How I work: the checks

Most of my skills are checks.

Each one is tested: it has to load when it's asked for and stay out when it isn't. The pictures are real output, not mockups. Pick one to look closer.

About

I build things to understand them.

I run MGKCodes, the studio behind frunt and MGKFitness. Away from the desk: gym, golf and gaming.

Contact

Hiring? Let's talk.

CV (PDF) · GitHub · LinkedIn · X · MGKCodes · mattykay2002@gmail.com