bjark.is

Full-stack and machine learning engineer. I build products end to end, from mobile apps and web to the backend and the ranking models inside them. Co-founder of Vestra AI, MSc Machine Learning student at KTH in Stockholm.

Bjarki Sigurjónsson Thorarensen

Now
Building Slapp at Vestra AI, MSc Machine Learning at KTH
Where
Stockholm , Reykjavík

Slapp

Currently my company Vestra is building Slapp, a social app for Iceland where every user signs in with the national electronic ID. Two developers built it: an iOS and Android app, a web client, an admin dashboard, a merchant portal and a shop, all on one Convex backend.

  1. 22,000+verified users, over 5% of Iceland's population, with no paid marketing since launch in early 2026.
  2. 5apps on one backend: iOS and Android, web, admin, merchant portal and shop.
  3. 532klines of TypeScript and 2,782 commits on main.
  4. 2developers, from the first real commit in December 2025 to version 2.6 in August 2026.

Five clients, one backend

Clients

  • Slapp appiOS and Android856 files, 257k lines
  • Webwww.slapp.is93 pages
  • Adminadmin.slapp.is42 pages
  • Tilboð portaltilbod.slapp.is15 pages
  • Shopmerch.slapp.is1 page, 4 routes
Convex
278
tables
262
modules
88
cron jobs
63
components

Database, functions, scheduler, crons, HTTP routes and auth in one deployment. Clients connect with a Better Auth session and a one-hour Convex JWT; services are reached by fetch or SDK, configured by 62 environment variables.

Services

  • KennieID sign-in
  • Cloudflare R2Images, audio, print files
  • Cloudflare StreamVideo upload and playback
  • Google Cloud KMSChat encryption keys
  • RevenueCatSlapp+ subscriptions
  • Expo PushAll notifications
  • ResendMerchant and school email
  • OpenAIAgent, news, parsing
  • Shopify and PrintfulTee checkout and fulfilment

Also fal.ai and Gemini Lyria for AI images and songs, API-Football, Google Places and Icelandic data sources.

Apart from two small prediction APIs for the football games, all of Slapp's own code runs in this one Convex deployment.

The case study

  1. 01SlappWhat Slapp is, where it runs, and the key numbers.
  2. 02IdentityThe eID identity model: no passwords, a verified birthday, and a member number.
  3. 03ArchitectureThe Convex backend, the five clients, and how data moves between them.
  4. 04ProductThe feature map and every major surface of the app.
  5. 05Machine learningEvery model is logistic regression or a self-trained embedding, trained inside Convex.
  6. 06MonetisationSlapp+, Tilboð offer slots, loyalty clubs, invoices and the member tee.
  7. 07ScaleRepository facts, commits per month, the release timeline, and size by area.

Coverage of the launch by Morgunblaðið, Vísir, Sýn, Viðskiptablaðið, K100 and Bylgjan is listed in the case study.

Read the case study

Machine learning

I am studying machine learning at KTH in Stockholm, and Slapp is where I put it into production. Nine models rank the feed, suggest friends, gate push notifications, predict churn and order the dating deck.

Every one is a logistic regression or a self-trained embedding, trained in TypeScript inside Convex and shipped only if it beats the current model on held-out data. The feed ranker retrains every night. The notification ranker halved push volume while keeping 87.6% of opens. The churn model's effect is measured against a 10% holdback group.

The machine learning chapter

Vestra AI

Vestra AI ehf. is the company behind Slapp, founded in Reykjavík in early 2025 by two technical founders, Kristján Leó Guðmundsson and me.

About Vestra AI, and the 30 ideas of 2025.

Writing

Veist þú í raun hvernig gervigreind virkar?

Do you really know how AI works?

Vísir, 3 September 2026

An explainer for a general audience on how large language models work (tokens, vectors, attention, weights) and where they fail, including hallucinations and the METR study of developer productivity. It closes with what a Statistics Iceland survey shows about the gap between AI use and AI strategy in Icelandic companies.

All articles.

Experience and education

  1. 2020 to 2023University of IcelandBSc Software Engineering, GPA 8.66 of 10.
  2. 2023 to 2025Juni DigitalReykjavík. Web developer: Next.js, TypeScript, GraphQL and headless CMS.
  3. 2025 to presentVestra AIReykjavík and Stockholm. Co-founder and co-owner. Slapp.
  4. 2025 to 2027KTH Royal Institute of TechnologyStockholm. MSc Machine Learning, expected December 2027.

Experience and education in full.

About

Bjarki Sigurjónsson Thorarensen

I live in Stockholm, where I study machine learning at KTH and build Slapp with Kristján. Icelandic is my native language, my English is fluent, and I am studying Swedish.

GitHub, LinkedIn, vestra.is, slapp.is, CV as PDF, more about me.