Who am I?

I am a backend and MLOps engineer — I design and run the systems that carry machine-learning models into production: real-time data and feature pipelines, low-latency inference services, and the deployment, monitoring and promotion machinery that keeps them reliable under load.

My work joins two lines that are usually pursued separately. The first is doctoral research on hardware-aware inference for vision–language models and vision transformers in safety-relevant driving-scenario understanding, carried out at the Elios Lab (DITEN, University of Genoa).

The second is eight years architecting production machine-learning infrastructure — real-time feature stores, low-latency inference gateways and MLOps promotion pipelines — at Snapp, Iran's largest ride-hailing platform, for a service with 50M+ registered users. I left as Senior Engineering Manager for Backend & AI Platform, having founded and led the 11-engineer team that owned it.

The thread through both is getting reliable intelligence out of scarce resources. It began with scarce labels — semi-supervised learning, and the Semi-CART method I published in Springer's IJMLC — and now runs through scarce compute: what survives when a model has to answer inside a latency budget, on a device, next to a person or a vehicle.