Krzysztof Duda

hello@krzysztof.studio

i’m a product engineer. i take rough ideas and ship them end to end, from the first interaction to the full-stack system behind it.

currently building Indeks and tools for working with AI agents.

email me ↗see the work ↓

Surveyor

describe the ride you want and the map draws it.

Next.jsMapLibre

0:000:00

IMAX Canvas × Jev by TypeSafe AI

describe a look in one line and the IMAX Canvas console restyles itself. there’s no LLM and no generated CSS, because TypeSafe AI’s Jev System One only picks from thirteen closed choices and then scores fifteen dials against them, so “a bit warmer” moves warmth and nothing else, for about $0.0005 a directive.

TypeScriptWebGL2Jev / TypeSafe AI

0:000:00

Indeks × Jev by TypeSafe AI

write one sentence about a collection and Indeks files the whole library into it while you watch. TypeSafe AI’s Jev System One answers every bookmark with a probability instead of a sentence, so above 0.85 it files, below 0.35 it drops, and only the middle comes back to you as a question. 1,132 bookmarks in under 40 seconds.

Ruby on RailsNext.jsJev / TypeSafe AI

0:000:00

IMAX Canvas

accessible Web Components dressed up as a late-1990s cinema projector console. plain TypeScript custom elements, no framework, painted by a WebGL2 amber phosphor shader: bloom, scanlines, curvature, grain, flicker.

Quorum

ask one question and a few research agents each take a different angle, in parallel, read-only and blind to each other. a judge reconciles their answers into one cited note, and notes build on each other across runs. native macOS app on top of the Claude Code CLI, with a spend cap and 319 tests.

SwiftClaude Code

domometr.pl

paste an Otodom listing and Domometr tells you whether the price is fair, with a report and a chat that answers in maps, tables and charts. the model never writes UI: it picks from 28 typed components through json-render, streamed over SSE from Rails and RubyLLM.

Ruby on RailsNext.jsRubyLLM

0:000:00

canonical_log

Ruby gem for the canonical log line pattern. it collects context over the whole request and writes one structured JSON line at the end, instead of log calls scattered everywhere.

Ruby

Visualizing bookmark clusters on a globe

bookmarks placed on a 3D globe by meaning. each one gets a vector embedding, an LLM called from Ruby turns similarity into coordinates, and related links end up next to each other.

0:000:00

Indeks

i had 2,000 bookmarks and had read maybe 50, so i built a place where they sort themselves. save an article in one click, find it with search, and let embeddings group and tag everything, no folders.