rene zelaya // one-person studio
Shipped Solo
a repeatable framework for the one-person studio in the age of AI
A repeatable system for taking an idea to a shipped product, alone, with AI models as the rest of the studio. The claim is about scope, not speed. Six phases. One operator.
the system
Six phases, idea to shipped
One move, repeated at the six places work with AI tends to break. The tightest version: restrain the human from the trap of the overconfident AI. It is a method, and it can be taught.
training
It is a method, and I teach it.
of organizations offer training for LLM competency. Everyone else is handing people powerful, overconfident tools with no method for using them well. That gap is the real risk, and closing it is the work.
Want to learn this methodology? I turn the six phases below into training and one-on-one coaching on the Shipped Solo Method, for individuals and teams alike, so you get a repeatable way to ship with AI instead of winging it.
Train with me →Calibrate
Test the AI where you are qualified to grade it, so you learn its edges before trusting it where you can't.
Research
Desk research as a sieve. Make the AI argue against the idea, and let most ideas die cheap.
Spec
Groom the prompt like a Jira ticket: the data, the intent, and the verification, composed into one directive.
Build MVP
Build only the core flow first. Get it right. Add features one at a time on top.
Test
Real users, fast. Their feedback decides whether you harden it, add to it, or kill it.
Interrogate
Question the output and the reasoning instead of accepting it. Make the AI defend the options it rejected.
the proof
The scoreboard
Real, live work. Open the product.
WhisperPad
Local, privacy-first dictation running Whisper entirely on-device. The flagship.
PixelPinch
A menu bar app that captures, downsizes, and compresses screenshots so pasting them into ChatGPT, Claude, or Gemini costs roughly 4× fewer tokens.
Transcribd
On-device diarization with speaker labels. No cloud, no upload.
IntentOps
A form for logging studio work: capture intent, reasoning, risks, what got done, blockers, and the AI’s role, then export a structured log. A hosted, authenticated web app (sign-in gated).
Skycritter
A pixel-art kitten weather app. Cosmetic-only extras, no paywalled features.
Sample Size Calculator
Plan study sample sizes without the spreadsheet gymnastics.
Significance Calculator
The A/B test calculator: compare success rates between two versions of a design and find out if the difference is statistically significant, in plain language.
flagship case study
WhisperPad
A macOS voice-to-text app that runs OpenAI's Whisper model fully on your machine. No audio leaves the device, ever. It ships in two builds: one on the Mac App Store and a direct build from the studio site, both with full auto-paste through the Accessibility API.
When Apple rejected the auto-paste feature, the writeup about it ran on the front page of Hacker News. Then Apple called and reversed the rejection. The constraint became the story, the story became the channel, and both builds now ship in parallel.
Read the full story on the blog: the rejection · the reversal.
the operator
About
I'm Rene, a senior UX researcher with eight years across SaaS, telecom, and financial services. I'm finishing a master's in HCI and AI at Elisava in Barcelona, and MITM LLC is the studio where I ship.
The argument behind all of it: AI doesn't just make a solo builder faster: it changes the composition of work one person can attempt, from single-threaded producer to orchestrator of capable but mistake-prone agents. The scoreboard is the evidence. The framework is how it repeats.
work with me
Relocating to the US in fall 2026. Open to UX research and AI product roles.
Also up for talking about solo shipping, on-device AI, or any of the work above.
