Shervin Rohani

At Cafe Bazaar, I built and improved machine learning systems for an Android marketplace serving 40M+ users. I fine-tuned transformer models for personalized app similarity ranking, increasing installation rates by 40% and delivering a 6% net lift in App Details installs.

I’ve worked across recommendation, search, advertising, and auction optimization. My work includes a Two-Tower neural network using TabNet that improved offline MRR by 70%, CTR estimation and debiasing models that improved Tune Factor by up to 35%, and auto-bidding algorithms that increased installation value per bidder by about 15%.

At Lendo, I built the company’s first end-to-end data infrastructure and credit-scoring system from scratch, including an automated pipeline for 60K+ loan records. I also reduced data-pipeline runtime by 12x at Cafe Bazaar by refactoring large-scale DAGs for incremental processing.

I enjoy bridging theoretical mathematics with scalable production code, using machine learning, optimization, and modern AI tooling to solve real business problems.