At Labelbox, Inc. (Alignerr), I evaluate and rank model-generated code for an AI-training project, assessing correctness, realism, and compliance with task specifications. I write structured justifications for each judgment.
On my KriptoFarsi platform, I audited an implausibly strong result and traced its headline +310-percentage-point figure to a sign-flip in the accounting code. I rebuilt the measurement layer as a correction-aware historical replay engine, which recovered a signed result of +54.45pp before realistic execution costs reduced it to roughly +1.8%.
After statistically testing 700+ recorded signals and finding the strategy indistinguishable from noise, I retired the operation rather than publish an inflated result. I also identified and fixed a phantom-entry bug that had corrupted the recorded trade set.
As an independent software engineer and consultant, I build and operate production Python services and Telegram bots, including admissions flows for TRstudy / UniKiani and a currency-exchange intake backend for Kiani Exchange. Earlier, at Alien Games, I owned production infrastructure for a live online game and handled uptime, incident response, and DDoS mitigation.

