PL FAST EN
AI / Product / Fullstack Engineer

Michał
Wiencek

Kraków, Poland · available for work

I build AI tools and interactive websites that people actually open: fast, responsive, and able to read between the lines. Eight of them are listed below. Every one has an address, not a screenshot from a deck.

This page is the fast version: flat, in English, everything on one screen. The two places where the rest of it lives are right below.

A drawn room: Michał sitting at a desk with two monitors, a corkboard on the wall and a shelf of project covers The real portfolio is a room you walk through. Built in three.js. The writing inside is Polish. Open it → And the library, where everything else is filed. The long-form CV, the candidate dossier, extended write-ups of every project and the things that did not fit in a room. Also in Polish. Open the library →

What I get hired for

Three things, and they overlap more often than not.

AI agents and LLM pipelines Multi-stage analysis, structured output contracts, tool calling over MCP, evaluation harnesses and an explicit point where a human takes the decision back.
Interactive web experiences three.js and WebGL scenes, scroll choreography, generative visuals. Built to hold 60 frames per second on a mid-range phone, not only on my machine.
Product engineering end to end Next.js and TypeScript, browser-side computation when the data should not leave the device, payments, deployment. From an idea to something with an address.

Selected work

Ordered by how much of me is in it, not by date.

  1. PodTeksT

    A conversation analyser for chat exports from Messenger, WhatsApp, Instagram, Telegram and Discord. Metrics are computed in the browser, so the conversations never leave the device. On top of that sits a multi-stage psychological analysis, a battery of tests and a semantic search engine with Polish NLP.

    Next.js · TypeScript · local embeddings · multi-stage LLM pipeline

  2. Janosik schronów, a shelter map of Poland

    86,591 shelters and emergency refuges from the State Fire Service register, each cross-checked against the building registry, the national topographic database, OpenStreetMap, the address registry and demolition permits. Reports from the street decide whether a door actually opens.

    Vector map · open data · Cloudflare

  3. okkropkastudio

    My animation studio. Its series Wyświetlono (Seen) is about a group chat everyone has muted and still reads: five friends, one-minute vertical episodes.

    Remotion · Blender · vertical video

  4. Parter, a shop that sells shops

    An online shop whose entire stock is other online shops. Empty ones, ready to be filled. The joke is the product and the product works.

    Web shop · custom storefront

  5. Latarnik Simulator 2026

    A browser game about a lighthouse keeper, written entirely in the front end. No server, no build pipeline for assets, no install.

    Canvas · vanilla JavaScript

  6. 59 days

    The documentation of a medical malpractice case, laid out as a readable timeline instead of a folder of scans. Built because the paperwork was unreadable.

    Long-form document · timeline

  7. A lecture from Sapolsky

    An interactive presentation on the biology of stress, following Robert Sapolsky's work. Scroll-driven, with the diagrams reacting rather than sitting still.

    Scroll choreography · SVG

  8. 30 synthetic Polish personas

    A corpus of Polish narrative biographies with an analysis pipeline, tested across seven language models. It is the dataset behind the publication below.

    Dataset · CC BY 4.0 · DOI on Zenodo

Research

One paper, one dataset, both public.

Wiencek, M. (2026)

Baseline Intercepts Versus Persona Slopes: Stimulus and Administration Fidelity of Polish Narrative-Biography Personas in Large Language Models

The question is whether a persona prompt actually moves a model's answers, or whether the model simply answers from its own baseline no matter who it is told to be. The data package and the analysis pipeline are released under CC BY 4.0.

Background

Where the psychology in the AI work comes from.

Psychology University of the National Education Commission, Kraków. Master's programme, fifth year.
Cognitive science University of the National Education Commission, Kraków. Three years.
Technical secondary school, IT Bielsko-Biała. Networks, systems, algorithms.
Overwatch World Cup Team Poland, 2018 and 2019, as DXTR. The youngest player on the roster.

Also

I run an eight hour one-to-one course on building AI agents and automations. It is taught in Polish; the full syllabus is online, all twenty three lessons of it. I also make films: 3D scenes from the PodTeksT world and three ad demos, twenty in total, all in one gallery. Polish is my first language, English is C1.

Let's work together

Write in English or in Polish, whichever is easier. I answer both.