An assessment should end as advice a person can use, not a score they file away.
I own that path. Retrieval sits under the agents and the scoring, in TypeScript, NestJS, and Python on
Kubernetes, so the model is reading that person’s self- and peer-assessment. A held-back answer comes back
as a line they can use.
A sentence, hidden in code, with no timer.
A clock would have turned the pattern into a race, so the solve stays open. Persian sentences sit behind a
letter-and-number code. I co-own it, and I lead the backend, the AI, and the infrastructure. It is on Cafe
Bazaar, with 26k downloads on Myket.
Dots and boxes, played live against someone else.
The game lives on a table, with a pencil. I co-founded it and lead the engineering. Unity on the client.
Node.js and MongoDB microservices underneath, self-hosted with Docker. It is on the App Store.
The servers behind a turn-based multiplayer game with a million downloads.
I took those servers and moved the live game
onto a new cluster
without taking it down. Microservices with gRPC under the matches. In the seventeen months I was lead,
logging and monitoring cut incidents by 40%, and the cluster held at 99.9%. The container work from 2021
to 2023 cut yearly server costs by 25%.