Ethan Hubbartt

Senior Software Engineer · New York

I design what I ship, and ship what I design.

I took Teleparty's mobile app from an idea to 1M+ downloads, then pitched, directed and built its redesign: my concept, our designer's Figma, my code in production. I now lead four engineers and a designer across Android, iOS and the browser extension.

1M+
downloads of the mobile app I built
~80%
less onboarding drop-off
8
streaming services reverse-engineered
~20k
users of my open-source plugin

01 · Side A

Teleparty: the redesign

Turning a single-purpose watch-party tool into a place people want to open.

  • Role Concept, product direction and all engineering
  • With One designer, in Figma
  • Timeline About 3 months, first prototype to launch
  • Stack Swift, Java, Liquid Glass, Material 3 Expressive
Walkthrough of the new appmedia/redesign-hero.mp4
The new app: home feed, choosing a service, starting a party
+16%
average hours in the app, week over week, three weeks running since launch
+15%
parties started, week over week, across the first month
+0.2
store rating since launch
~80%
less drop-off in initial onboarding after the sign-in rework

Early results: the redesign is weeks old, so these are trends, not settled numbers.

Why it happened

Our metrics showed people only opened the app with one goal in mind: start a party. They did that and left. We also saw an extreme drop-off during onboarding, where the flow had too much friction. I pitched a redesign to fix both.

My role

The redesign was my idea, and I decided most of what it would do: the features, how they would work and how the app would look. Our designer turned that into Figma files I could build from pixel for pixel. I did all of the engineering.

A design change that deleted code

The old app had to manage a sign-in state for every streaming service we integrate with. In the new design, people handle their own sign-in with each service, so the app no longer tracks it. Onboarding lost its biggest source of friction and the codebase lost a whole class of state.

The hardest decision

How people start a party and switch between streaming services. It is the core of the app and something existing users know by habit, so the new version had to be easier for newcomers without confusing anyone who already knew the old way.

Built to keep growing

We launched about three months after the first prototypes. The design is modular, so new pieces ship as they are ready instead of waiting on one big release.

How it feels

A personalized home feed of suggestions, built-in tutorials and several clear routes to invites, in iOS Liquid Glass and Material 3 Expressive so each platform feels native.

Before and after

Afterredesign-home-after.jpg
Beforeredesign-home-before.jpg
BeforeAfter
First screen you see: the old Browse tab against the new Home
Afterredesign-party-after.jpg
Beforeredesign-party-before.jpg
BeforeAfter
Party: drag to compare before and after

What we didn't ship

First prototypemedia/redesign-rejected-1.png
Home, first prototype

The very first prototype

This came out of an early ideation session on what the home screen could become. The core idea survived: a home page with your current party and things to continue watching. Two parts of it did not.

Browse in the centre of the nav

We wanted Browse to be the centrepiece of the navigation bar, but never found a clean way to make it work. In the shipped app it is a regular tab alongside Home, Party and Profile.

Hero art that wasn't us

Leaning on film artwork for the hero felt a little tacky and pulled away from Teleparty's own branding, so the final design stays closer to the brand.

Late prototypemedia/redesign-rejected-2.png
Home, one of the last prototypes

One of the last prototypes

The opposite end of the process. In its best form this looks amazing, arguably better than what we shipped. That was the problem: it only looks like this in its best form.

Design for the worst state

A design has to look great at its worst, not only at its best. This one puts a streaming service's poster across the whole home page, and we don't control those posters: their orientation, aspect ratio and resolution vary from service to service.

Keeping control of the home page

We couldn't guarantee it would look good for every user, and it took the quality of the home page out of our hands. We chose the design where we could ensure a good experience every time.

02 · Side A

Teleparty: under the hood

From an idea to 1M+ downloads: two years as the only mobile engineer, then building the team.

  • Role Sole mobile engineer, then team lead
  • When 2022 to now
  • Stack Java, Swift, Jetpack Media3, Frida, WebViews, AWS

Synced playback on 8 streaming services

I reverse-engineered eight streaming integrations on my own, including Netflix, Paramount+, Hotstar and Crunchyroll, and recreated synced playback natively on mobile.

I recently took our Netflix error rate from 50% down to 4%. Most of what remains is Netflix's own limits, such as too many screens on one account or a title not being available in a region.

Fixes that ship in minutes, not days

I came up with and built a config-serving platform that injects versioned scripts into iOS and Android WebViews. When a streaming service changes, the fix goes out in minutes instead of waiting days on store review. It has repeatedly fixed problems for tens of thousands of users that way.

It also has live logging built in, so I can watch one person's session and fix their issue in real time.

AI agents testing on real hardware

I built a device-testing workflow on Claude Code that lets AI agents exercise the app on real Android phones.

Building the team

I ran the entire hiring process for the four engineers and the designer I lead, and I interview for roles across the company, including teams I don't manage. Day to day I plan the work, review code, train and mentor, and unblock people.

1M+
downloads, from zero
50% → 4%
Netflix error rate
Minutes
to ship an integration fix, down from days

03 · Side B

Volition

An event platform for an Old School RuneScape clan, where in-game drops claim tiles on live game boards.

  • Players About 400 in a single event
  • Stack SvelteKit, TypeScript, Supabase, Cloudflare Workers, three.js
  • Link volition-osrs.com
A board updating live as drops come inmedia/volition-board.mp4
Replay of a clan-vs-clan Connect Four event: in-game drops claiming a shared 600-tile board, with the live drop feed below

What it is

A game on top of the game. Discord sign-in, an admin event builder, and interactive boards: Battleship, bingo, and clan-vs-clan Connect Four on a shared 40×15 board. Players upload images and play against each other as it happens.

The live pipeline

A Cloudflare Worker ingests RuneLite drop webhooks and triggers crediting. Versioned polling keeps every board live for hundreds of players at once, so a drop in game claims its tile without anyone touching the site.

The fun parts

3D card-pack reveals in three.js. I also took over the clan's Discord bot and rebuilt it with member verification, tickets and points.

Home pagemedia/volition-home.webp
Home: events calendar, member stats and ranks
3D card-pack revealmedia/volition-packs.mp4
three.js pack reveals
~400
players in a single event at peak
100,000s
of in-game drops processed
Dozens
of events run on the platform

04 · Side B

Combat Achievements Tracker

A RuneLite plugin for Old School RuneScape with around 20,000 users.

  • Users ~20,000
  • Stack Java, RuneLite
  • Link GitHub

An in-game UI for tracking Combat Achievements. Tier thresholds are read dynamically from game state, and completion rates are pulled from the wiki, so it stays correct as the game changes.

Along the way I contributed a merged pull request to RuneLite itself, the open-source OSRS client.

The plugin panel in gamemedia/tracker-ui.png
The plugin panel inside RuneLite

05 · Side B · In development

Goldfish

Full games of Magic: The Gathering Commander, played entirely offline. No existing MTG app offers that.

  • Status In development
  • Built Rules engine and AI opponents, from the ground up
  • Engine ~12,000 of Magic's ~30,000 cards
A turn being played against the AImedia/goldfish-game.mp4
A three-player game against two AI opponents, on iPad

What it is

A Commander app with its own rules engine and AI opponents, plus solo playtesting ("goldfishing") and a deck analyzer.

Where it stands

The engine handles about 12,000 of the roughly 30,000 cards in Magic, including around three quarters of the most-played Commander cards. The app itself is close to finished; the engine is where the work continues.

Liner notes

About

I'm a senior software engineer at Teleparty in New York. I joined in 2022, built the Android app from the ground up, and spent two years as the only mobile engineer. Before that I built computer-vision and OCR tools at Gets USA that replaced manual damage assessment of computer components.

I care about the whole path from a Figma frame to a shipped screen, and I like owning both ends of it.

Outside of work, it's music. I think it's the most important art there is. I grew up playing piano, guitar, trumpet and saxophone, and I'll listen to almost anything, with Radiohead, Phoebe Bridgers and Kendrick Lamar near the top.

BA, GIS: Data Science · University of Washington

Languages
Java, Swift, TypeScript, JavaScript, Python
Frameworks
Android SDK, Jetpack Media3, React, SvelteKit, Next.js, three.js, Node.js
Design
Figma, iOS Liquid Glass, Material 3 Expressive, accessibility
Infrastructure
AWS, Supabase, Cloudflare Workers, Fly.io, GitHub Actions
Specialties
Reverse engineering (Frida), WebView script injection, AI agent tooling