AI engineering at MGX
Building applied AI systems with a bias toward reliable interfaces and useful loops.
John Jung
I’m John. I build AI systems and small, useful products.
My work has moved through cosmic-ray data, clinical bioinformatics, email infrastructure, and now AI engineering. The thread is pretty consistent: understand the system, make the interface clearer, and ship something people can use.
Currently working on AI engineering at MGX and keeping a small lab of projects across developer tools, productivity, finance, and agents.
Now
Current work, kept simple.
Side-project lab
Developer tools, productivity, finance, and agents.
Interested in
Useful agents, reliable AI systems, product craft, and small tools.
Timeline
A restrained archive of the path here.
The through-line has been data, interfaces, systems, and shipping useful things.
Early work
NASA Goddard / University of Maryland
Built and analyzed cosmic-ray data systems, where the work was mostly about signal, noise, instruments, and patience.
Clinical data
NYU Langone / University of Rochester
Graduated at 19, then spent several years on clinical and research bioinformatics datasets, contributing to 30+ publications.
Product studio
Pith Studio
Built product work for startups and larger companies, including Fortune 500 client work, with a focus on making rough ideas usable.
Email systems
june.ai, then Nylas
Started june.ai, an AI-first email app acquired by Nylas, then worked on email and productivity infrastructure at Nylas.
Now
MGX
Working on AI engineering and keeping a small shelf of practical experiments moving.
Selected Projects
A project shelf, not a trophy case.
A few useful tools and experiments from the lab, written as notes on what they are and why they exist.
Webhook Rodeo
webhook.rodeo
ActiveDeveloper tools
Webhook infrastructure, but with a better debugging loop: receive, transform, route, and replay events.
Why: Webhooks are everywhere, but debugging them still feels worse than it should.
Receptive
withreceptive.com
ActiveAgents
An AI answering service for catching missed calls and turning them into useful follow-up.
Why: Phone calls are still where a lot of important work slips through.
Stride Systems
stride.systems
ActiveAgents
AI agents for operational follow-through across tickets, docs, and team pings.
Why: The value is in closing the loop after the model produces a suggestion.
skeet.build
skeet.build
ExperimentDeveloper tools
A small bridge for connecting Cursor to MCP tools.
Why: Good agent tools should make context feel local and easy to wire.
check.supply
check.supply
ActiveFinance
A mobile-first way to send paper checks without envelopes, stamps, or post-office errands.
Why: Old financial rails still need clear interfaces.
june.ai
june.ai
AcquiredEmail
An AI-first email app, later acquired by Nylas.
Why: Email is a messy coordination system; the question was how much of it could feel calm.
Lab Archive
Smaller experiments, still part of the shelf.
Collective
Shared budgeting, expense management, and spending controls for groups.
ExperimentOpenMyWalletPass
Turns membership, library, park, gym, and club cards into Apple Wallet passes.
Weekend buildOpenBuoy Club
Social listening for creators, founders, and teams in the Farcaster ecosystem.
ExperimentOpen
Notes
A few working notes.
Short entries about building reliable AI products, small tools, and the path from research data to product systems.
What makes an AI product feel reliable?
Reliability in an AI product is less about sounding certain and more about showing the user where the system is, what it knows, and what happens next.
Notes on building small tools
Small tools are useful because they can stay close to one problem, one loop, and one kind of relief.
From cosmic rays to product systems
A short note on the common thread between research data, bioinformatics, email infrastructure, and AI product work.
Principles
A small operating list.
- Useful beats impressive.
- Ship, then refine.
- Systems over demos.
- Taste is care.
- Stay close to the material.