Sebastian Rücker

I build AI engineering teams and can fill any seat on them, human or AI.

sebastian@ruecker:~$ open full_cv

Facts

  1. 12 months from the first hire to an AI stack in production on the firm's own hardware. Head of AI at Lang & Schwarz, a market maker regulated under DORA.

  2. 30 people: the largest team I led while writing code myself, at Optiopay and Compredict.AI.

  3. 59% profitable trades after costs, out of sample, against a 39% baseline. The model I built predicts price shocks from real trading data and social media sentiment.

  4. 1 gateway I designed: every AI agent and workload at the firm runs through it, among them a company-wide assistant and an EU AI Act training agent.

  5. 31 minutes: a planted regression on my own agent factory went from alert to a fix in production with no human.

Projects

AI Software Factory

Source available.

Agents build software from a spec. A human steps in only where it matters.

Agents build software from a roadmap, card by card, each in its own sandbox on your servers, and nothing merges until the checks pass. On the reference cluster a planted regression went from its alert to a fix in production in 31 minutes with no human, the factory fixed five of its own open defects, and one spec card became ten planned cards and a deployed link shortener.

Open aisoftwarefactory.worldCode
AI Software Factory console, chat and board of agent tasks

Hired

Live. Coming to the App Store.

Practice job interviews in realistic AI video calls.

Interview practice in realistic AI video calls, built as a white-label React Native app with Firebase and a separate backend. The iOS release is in review and Google Play is next.

Open tryhired.com
Hired landing page

EU AI Act training as an adventure-game dialogue. A Socratic AI tutor, then a test.

A Socratic tutor that teaches the EU AI Act from the Act's own text, coaches without giving answers away and runs a ten-question test. Python and LangGraph behind a LiteLLM gateway, a React front end in a pixel-art adventure dialogue, German first.

Try the demoCode
AI Act Trainer dialogue screen

Cannonade

Play in the browser

Build walls, enclose land, fire cannons. Siege strategy in the browser.

A remake of the 1990 arcade classic Rampart: build walls, enclose land, fire cannons. Godot 4.7, data-driven buildings with hot reload, AI opponents and a spectator mode, ported from my earlier Unity version.

Play in the browserCode
Cannonade: two castles trading cannon fire beneath the game logo

Contact

Consulting requests: consulting@sebastianruecker.com

Download PDF