At verido.ai, I hired and led a 10-person team and set technical direction for an AI identity-verification platform. I replaced manual document checks with an automated pipeline, cutting customer onboarding from 24 hours to 5 minutes and powering 150K+ successful verifications.
On Zehnlab’s Soro project, I owned the training data for the first large language model for Tajik: a 100M-token corpus and 10M tokens of synthetic task data. Soro ranked #1 on a curated Tajik benchmark across six subjects. I’ve also shipped production risk models used by 50+ banks and microfinance institutions, and my first-author research improved frozen OCR performance on degraded images.
