Side Projects

Side projects, mostly AI

Everything here sits outside the client work. Most of these are Claude skills and agents I wrote for problems I kept hitting in my own process, along with the products some of them grew into. A few study projects from MOME sit at the end.

AI skills and agents

Skills and agents I wrote for my own work. Most of them started as a task I was repeating by hand until the rules were worth writing down.

  • Skill

    voice-applier

    2026

    A Claude skill that strips AI tells out of prose and re-casts it in a specific person's register. It works from a pattern catalog and a tiered lexicon, so it restructures the writing instead of swapping synonyms.

    Why it existsI kept rewriting generated text by hand to make it sound like me, and I was making the same corrections every time. A skill does that pass in one go and applies the corrections consistently.

  • Skill

    cv-voice-applier

    2026

    An extension of voice-applier for CV prose. It adds a banned recruiter vocabulary, the structural tells specific to CVs, and a rule for replacing vague passion claims with something checkable.

    Why it existsRecruiter-facing writing has its own set of tells, and a general humanising pass does not catch them.

  • Skill

    cl-voice-applier

    2026

    The cover-letter layer on top of voice-applier. It bans formulaic openers and closers, catches recruiter-detectable structure, and requires a personal anchor in place of a generic motivation line.

    Why it existsCover letters fail in a narrower set of ways than CVs do, and almost all of them are in the first and last paragraph.

  • Agent

    job-searcher

    2026

    An agent that finds open roles across job boards and company career pages, then filters them against a candidate profile instead of returning everything that matched a keyword.

    Why it existsThis was the first of the three terminal scripts that eventually became JobEmber.

  • Agent

    employer-profiler

    2026

    An agent that researches a company before an application goes out, covering what they build, how they are funded, who works there, and what their design function looks like.

    Why it existsI was doing this research manually for every application and producing the same shape of notes each time.

  • Agent

    cv-adapter

    2026

    An agent that reshapes a CV against a specific posting, reordering and reweighting real experience without inventing any of it.

    Why it existsTailoring a CV is mechanical work with a hard constraint, since nothing in the output may be true of a different person.

  • Skill

    ai-detector

    2026

    A screening pass that scores generated application documents for the patterns detection tools look for, and reports what to change before anything is sent.

    Why it existsWriting in someone's voice and passing a detector are different problems, and I wanted to see both scores.

  • Skill

    position-relevance-filtering

    2026

    A scoring skill that evaluates a posting against a candidate's CV and target roles, returning a hard-filter verdict plus a relevance score from 0 to 100.

    Why it existsA job search produces far more postings than anyone can assess carefully, so the triage needed to be consistent above all.

  • Agent

    ux-wise-agent

    2026

    A design-review agent that critiques an interface against usability and accessibility criteria, and reports findings with the reasoning attached, so each one can be argued with.

    Why it existsI wanted a second pass on my own work that would argue with me instead of agreeing.

Experiments

Products nobody asked for, built to find out how far AI tools can carry a whole design process.

  • FoodCare

    2026

    A recipe app for households where one person cooks around another person's diagnosis. I ran it as an imaginary product, start to finish, to find out how far I could take an app with AI tools alone.

    Why it existsI wanted to see what the whole arc feels like when AI carries the mechanical parts. So I invented a client, ran real interviews, and took it through to 18 high-fidelity screens and a prototype you can click. No client was waiting on it, which is why I kept going past the point where a paid project would have stopped. I enjoyed the whole thing more than I expected to.

Study projects

Work from the MA at MOME Budapest, including two briefs that came from companies.

  • K&H Bank personal finance

    2023

    A personal finance concept for K&H Bank, covering cash-transaction tracking, spending insights and a challenge mechanic. Run as an interdepartmental student project at MOME Budapest.

    Why it existsThe brief came from the bank directly, and the interesting constraint was cash. Hungarian household spending still runs heavily on cash, which no automatic categorisation can see.

  • Mercedes-Benz Gesture and Voice Control

    2023

    An in-car interface controlled by gesture and voice, designed with Mercedes-Benz as an interdepartmental brief at MOME Budapest.

    Why it existsMercedes-Benz gave us the brief, and we went looking for ways to make the interaction inside the car better. We started the work by interviewing drivers. The design came out of what they told us, and we presented it back to Mercedes-Benz.

  • Purify

    2023

    An app concept for people on long acne treatments, built for the weeks when they want to quit because their skin got worse or dried out badly. I made it at MOME Budapest with Iryna B, Polina V and Tetiana T.

    Why it existsThe medication takes months to show anything. We built the app around progress tracking and an assistant that answers questions at the moment they come up.