Mert Safa Kenar

2026

Istanbul: The First Night

On the first night after a major earthquake, more than one in four people outside cannot find space in an assembly area within walking distance.

2.6 millionpeople find no space within 1 km. Out of about 9.1 million people outside on the first night, median scenario.

I merged more than 25 open data sources into one table covering Istanbul’s 959 neighbourhoods and built a linear programming model that assigns people to nearby assembly areas by capacity. I did not estimate deaths, because that cannot be validated with open data.

Hover over the points: each one is a population block of about 250 metres. The deeper the orange, the larger the share of people in that neighbourhood who find no space.

Python, DuckDB, HiGHS LP, LightGBM, MapLibre, deck.gl

2026

Discovery Dynamics

More than half of the indie games on Steam go almost unnoticed in their first year, and copying a successful game does not change that.

54%of indie games get fewer than 10 reviews in their first year.

I studied 92,492 games released between 2018 and 2025. I wrote down the hypotheses before looking at the data, then tested them with methods such as matching and regression discontinuity.

Hover over the points: each one is a game, and similar games sit next to each other. The brighter the point, the more reviews the game got in its first year.

Python, DuckDB, LightGBM, UMAP, React Three Fiber

2026

market-queue-lab

When checkouts get busy, a separate line for each checkout keeps customers waiting less than one shared line.

65%checkout occupancy above which separate lines pull ahead.

A supermarket simulation that learns arrival and payment times from real checkout records. It is validated against queueing theory and against a period the model never saw, and it can be watched in the browser as a 3D store.

Python, TypeScript, React Three Fiber

2026Demo

BiteTrust

Someone with a food allergy should be able to see at a glance what they can’t eat on the menu.

An allergen-first QR menu for restaurants. The customer scans the code on the table, the menu flags the dishes that are risky for them, and the order goes live to the kitchen screen. The restaurant side also covers stock, reservations and shifts. It is currently a demo.

TanStack Start, Supabase, Cloudflare Workers