About Kolosei
Kolosei is a bootstrapped, founder-led studio building AI-powered Hebrew language tools — and the kind of privacy-first software that solves specific problems well rather than generic problems approximately.
Today that's two shipping Hebrew products and one health-AI research vertical. The thesis across all of them is the same: keep the user's data with the user, use the cloud only for heavy compute, and let a frontier model — Google's Gemini — do the work that genuinely needs one.
What we build
LinguistPro is our flagship — an offline-first Hebrew learning workspace. Its architecture is unusual: SQLite WASM running in browser-local OPFS, so the app delivers server-scale functionality without ever holding user data on a server. Gemini, Cloud Text-to-Speech and Cloud Translation run under the learner's own key. It's preparing for ulpan-cohort testing as part of a Hebrew-pedagogy diploma project.
HDLE Premium shares that Hebrew-NLP core but targets professionals — a desktop workbench for lexicography and terminology extraction, with encrypted credentials and Google Cloud Translation v3.
EgorGenom is a second, mission-driven vertical: parent-driven whole-genome rare-disease analysis, run in parallel with a clinical team and built on the open-source nf-core pipelines. It's research, held to clinical-grade methodological standards.
Founder
Kolosei is founded and run by Peter Kolosei — a developer and language-pedagogy researcher. The studio's bias is thorough over quick: enriching, complete analyses over subjective top-N shortcuts; live databases over stale local ones; reproducibility over cleverness; documentation updated in the same session as the work it describes.
That rigor is on the record. LinguistPro's research methodology is preregistered at OSF, and the genomics work follows a blinded multi-expert review protocol with a full reproducibility layer.
How we build
The default stack: Google Cloud (Gemini, Cloud TTS, Translation v3) for AI and heavy compute; Node.js + vanilla JS on the web (LinguistPro is intentionally framework-free for debuggability); Python + PyQt6 for desktop; Nextflow + Docker on WSL for bioinformatics; Astro for static content like this site. Everything ships on a containerised Hetzner + Coolify pipeline.