Ogün Gündogdu · AI engineer · Stockholm

I build AI apps
and I ship them.

I build LLM products with tool calling, guardrails and tests, and run them in production. Before that, seven years shipping games to millions of players. Everything below is live, so go and break it.

Try Sergeant Pace ↗ See what I've shipped
0players reached
0prototypes built
0launches into iOS US top charts
0apps live on the web today

Flagship

Sergeant Pace

An AI running coach that yells at you, plans your runs from couch to marathon, and is quietly terrified of you getting injured.

Knee hurts, stopped early. Effort 9.

Are you trying to retire early?! Effort 9 is way over your limit, and you only finished a quarter of the work. Stop running until that knee pain clears up and see a doctor if it persists.

log_runadjust_plan(step_back)

A real reply from the live app. The plan stepped back to shorter jogs and longer walk breaks, decided by the model but allowed by the code.

The model talks, the code decides

Gemini handles personality and judgment calls. Every number and every safety rule lives in plain, tested code. The model can't write a number into the plan.

Limits that slide, never jump

Effort and pace ceilings tighten smoothly with body fat and age. 29.7% and 30.1% get almost the same rules. Weigh-ins loosen or tighten them as you change.

Tested like production

124 tests, no API key needed. I broke the guardrails on purpose, ten different ways, to prove the tests catch it.

Operated, not just demoed

One recruit per visitor, per-IP rate limits, daily spend caps, error details hidden, push-to-deploy, one JSON log line per turn.

  1. Browser
  2. Flasksessions · rate limits · daily caps
  3. Gemini coach6 tools · retry · model fallback
  4. tools.pyplan · limits · safety flags
  5. Stateone file per visitor

What broke once real people used it

  • Visitors killed each other's requests. Each conversation made its own Gemini client and overwrote a shared one; when two people arrived together the first was garbage collected mid-request. Found it within minutes of going live. Now one shared client, with a regression test.
  • New visitors waited 70 seconds. On a free-tier key the top models run out of daily quota, and every new conversation retried them all. Now it remembers the last model that worked. Second visitor: 9 seconds.
Try it live ↗ Read the code ↗

It runs on a free-tier Gemini key with a daily budget. If the sergeant is off duty, the day's budget is spent. Come back tomorrow. It asks for your consent before using any numbers you enter and keeps them for 14 days at most. Read the privacy notice.

Shipped 4

Small, sharp tools, all live. Click through.

Before AI: games

Before AI, seven years in mobile games. The two I'm proudest of are on the App Store, and my own experiments live on itch.io.

The path so far

  1. AI engineer

    Building and shipping LLM products end to end: tool calling, guardrails, tests and deployment on my own server. Sergeant Pace is the flagship, with web apps like SL Live alongside it.

  2. Lessmore · AI-powered prototyping

    Paired with a game designer to turn ideas into playable casual-game prototypes. I used AI heavily across the whole prototyping loop, which is where it became part of my daily workflow, and why ideas got to a playable state, then kept or killed, so much faster.

  3. Phoca.io · co-founder

    Co-founded and led a startup studio. Built 50+ projects and put 3 of them into the iOS US top charts. Tap on Time passed 3.5M players and still has about 50K daily.

  4. Gameguru

    A new hyper-casual prototype every week: build it, test it, move on.

  5. Ruby Game Studio

    Part of 30+ commercial projects, from rough prototypes to full casual games. Played a major role in several of them, including Doctor Care, Drink Inc., Spa Master and Transport Master, all on the market.

  6. Margen Tech · VR

    Built and maintained a maritime simulator for VR in Unity, used to train and examine maritime personnel.

Need something built?

I'm open to collaborations and taking on client projects. Got an idea, a prototype that has to become real, or an AI feature you're not sure how to build? Tell me about it.

  • AI features and agentsLLM apps with tool calling, guardrails and tests, built to run in production.
  • Web apps and toolsFrom idea to a live app on your own domain, like Sergeant Pace and SL Live.
  • Games and prototypesPlayable prototypes fast, and polished mobile games.

Email me what you're trying to do, even if it's rough, and I'll tell you honestly whether and how I can help.

LinkedIn ↗ GitHub ↗ itch.io ↗