I’m a senior data scientist living in Rome, originally from Istanbul. I’ve spent the last 7+ years building data systems that actually run in production — first in telecom AI, now across mobile gaming and subscription apps that reach a billion-plus users.
What I actually do
At TapNation, I lead the analytics work behind the company’s mobile-game portfolio: a cohort-based LTV prediction system used by 100+ partner studios, a real-time bid and ROAS optimization engine plugged into the major ad networks, Marketing Mix Models that quantify organic vs. paid revenue, and most recently an LLM-powered Text2SQL system so non-technical teams can query their own data without paging the data team.
At Wavera (Famefy), I built the subscription LTV prediction stack from scratch — power-law renewal curves, EMA trial-conversion forecasting, late-conversion multipliers — so the marketing team can compute pROI within days of a user installing, not months.
Before that I spent four years at P.I. Works building telecom analytics for Beeline, Telefónica, and O2 — automated reporting pipelines, radio parameter optimization, the kind of work where one missed kpi means real people lose service.
The technical thread
I keep getting pulled into the same shape of problem: there’s a business decision that depends on a long-horizon prediction (will this user be worth the spend in 6 months? Will this channel still pay back next quarter?), and the data you have is short, noisy, and confounded. Most of my work is variations on this — fitting honest models to messy data, expressing uncertainty rather than hiding it, and building systems that other people can keep running after I leave.
Lately I’ve been spending more time on agentic AI. My MSc thesis at Sapienza was on agentic AI for 5G/6G telecom parameter configuration — RAG plus a multi-agent pipeline grounded in operator documentation. The consultancy work I’ve done outside TapNation (Text2SQL for restaurant ops at Backlogix, exam generation for EdTech at Pingula) is in the same space. I think the next two years of “ML in production” are going to look very different than the last two, and I want to be working on the change rather than catching up to it.
Outside work
I play semi-professional football, run when the weather lets me, and have an unreasonable interest in vintage things. I speak Turkish (native), English (C1), and enough Italian to make my Roman neighbors laugh.
Available for select consulting
I take on a small number of consulting engagements when the problem is interesting and the team is serious — typically around LTV/subscription modeling, marketing analytics, or LLM-powered data systems. If that sounds like you, reach out.