Rimeworks LLC · Roblox development studio
I build Roblox games and the infrastructure underneath them — using an AI development pipeline that turns weeks of work into days, with verification that runs against the real game instead of trusting a green test suite.
What I build
Every engagement runs on build slots — two weeks of focused delivery. Take one, or take several in parallel. What fills the slot is up to you.
Design brief in, finished Roblox title out. Built through the AI pipeline, tested against the running game, published when it works.
Four weeks of content drops, fixes and iteration on a live title — keeping a game earning while the next one is being built.
The infrastructure most studios never get around to: analytics, telemetry, deployment pipelines, internal tools.
Process
The speed comes from agents doing the typing. The reliability comes from refusing to believe them without proof.
The brief becomes a written spec and a staged build plan before a line of code exists. Ambiguity is cheaper to kill here than anywhere else.
AI agents implement against that plan, in parallel, with every architectural decision recorded as it is made.
Automated probes drive the real game — joining, buying, saving — and report what they actually observed. A passing unit test is not evidence.
Published behind a checklist, with a ledger showing which behaviours were verified and which were not. You see the same numbers I do.
Proof
Case study · shipped September 2026
A complete Roblox title taken from a handed-over design to a published game: lobby and matchmaking, a full VHS sorting loop with physical piles and shelf logic, saves, progression, gamepasses, mobile controls, leaderboards and analytics funnels. One developer, running the pipeline.
A years-old codebase too tangled to maintain, with retirement already being discussed. Rebuilt from the ground up with agents: every behaviour in the live game catalogued into 703 requirements, each one traced to the code it came from, then verified by probes driving the real place.
Reading every script to port it doubles as an audit. That one turned up 18 exploitable holes and over 300 behaviours silently broken in production — including a year-old backdoor nobody knew was there.
An event pipeline dropped into a studio's live games, reaching the titles that hold roughly 96% of its concurrent players — specified in an afternoon, with the implementation handled by agents.
The part that matters
A green test suite is not evidence that a game works.
AI writes code faster than anyone can review it. That is the selling point, and it is also the trap: a hundred thousand lines of passing tests can sit happily on top of a game that does not start, because every test built its own little world and never touched the real one.
So I build the checks that can actually fail. Probes that join the running game, press the buttons, buy the product, save the file, and report what they saw. A requirements ledger that refuses to call anything done until a named probe has watched it happen — and fails the build when a claim has nothing behind it.
It is slower than shipping on vibes. It is the reason a two-week build is a delivery date rather than a hope.
Systems
Studios usually know they need these. They rarely have anyone with the time to build them.
Drag-in event collection — sessions, chat, positions, custom funnels — landing in a database you own, with a read API and nightly rollups.
A web viewer that rebuilds the game's own geometry and paints player density over it, with recorded chat placed in world. Hot spots and dead zones, visible in seconds.
Whole-portfolio metric exports on a schedule, with data quality treated as a feature: unresolved days flagged rather than quietly reported as fact.
Rojo and Git workflows, CI/CD into QA places, and shared systems packaged once and dropped into every title that needs them.
Start here
Briefs, questions, or a game that needs rescuing — all welcome. I will tell you honestly whether it fits in a slot.
jake@rimeworks.llc