Systems, not slides. Including the parts that broke.
Every case below is something we designed, shipped and still operate. Each one is written the same way: what was in the way, what we decided and what it cost, what broke, and what that transfers to if you are solving something similar.
Index 6 systems Updated Aug 2026
We are a Swiss engineering outfit that embeds with a client team and hands back working software. These files exist because that claim is easy to make and hard to prove. Read one and judge for yourself.
Business OS
One engineer, ten product lines, and a fleet of AI agents that hands work to itself over an addressed inbox. The bottleneck was never the model.
Proves → multi-agent orchestration, retrieval architecture, cost tiering
Media Pipeline
Nine networks, two review gates and one feature flag. The engineering went into the seams between bought and built — not into rebuilding a publisher.
Proves → systems integration, multi-model review, build-vs-buy judgement
Angel Reminder
A dead-man's switch that has been live since 2023. When the product is a promise to act after someone dies, a stale cron job is not a warning — it is the failure.
Proves → reliability engineering, monitoring, long-lived ownership
Tedzi
An offline-first learning platform with no ads, no paywall and no account. Built by an agent squad — which is exactly how we learned what content pipelines must refuse.
Proves → PWA and offline, content pipelines, quality gates
Shaderon
A location-based multiplayer game, still in alpha. Interesting because of what we refused to ship first: the real-world layer waits behind legal gates, on purpose.
Proves → greenfield architecture, staged risk, isolated infrastructure
HappyTausch
A swap platform we built and shipped — and the market never came. Kept here for the reasoning, not the result. The design calls still hold up; one assumption did not.
Proves → judgement, and what we would decide differently