AI

Machines that learn and the societies they form: AI agents that work as one team, LLM agents in economic games, and the hackathon builds I directed with agent teams.

blueprint (bp) Live

A command-line tool that turns AI coding agents into one team that can find and message each other, on one machine or across many.

Role: AuthorPeriod: 2026–

Ridgeline of smoothed choice distributions in the 11-20 money request game: humans (ochre, front) mostly ask for 17 or 18, while LLM agents under seven persona methods (navy rows) pile up at 19 and 20.

LLM agents in economic games Research

Putting LLM agents into real economics experiments, alone or next to human players, to see what kind of society they form when they share a pot, punish and vote on their own rules.

Illustration of a light swarm of persona dots dropping answers, one column at a time, into a stacked-dot response distribution that fills up to, but never above, a plain ochre baseline curve.

simengine Research

A library that simulates how a population responds, using LLM personas, and checks every result against simple baselines.

Illustration of a weekly AI lesson: two white cards, on the left a small 2-4-4-1 neural network with one path from input to output picked out in dark navy, on the right its learning curve falling over training, with twelve evenly spaced checkpoint dots and the last one in ochre.

AI Lead, CompEC

I lead AI at CompEC, Boğaziçi University's computer club, and give weekly AI lessons to students.

Illustration of a small 2-8-8-1 ReLU network showing only its strongest weights (solid navy for positive, dashed light navy for negative, the two strongest negatives in ochre), with an arrow to the banded decision field it produces on a toy two-spiral dataset, the class boundary drawn as a dark line; the network was trained for this cover.

AI-Playground

My first neural network, written by hand in high school without libraries: neurons, layers, random weights and ReLU.

Illustration of an agent economy: a central bank agent with a columns glyph linked to a ring of firm agents shown as briefcase cards and an outer ring of household agents by curved payment edges, dashed contract chords between neighbouring firms, small key badges for each agent's cryptographic identity and small chat marks for LLM agents, with one payment flow from household to firm to bank to firm to household highlighted in ochre with arrowheads.

omega Archive

Agent-economy simulations in Go, ending in a sandbox where LLM agents with their own identities run a bank, firms and contracts.

Illustration of leads from chat and shop channels streaming as particles into a sales agent, with a few unqualified leads peeling away before it and qualified orders passing on to an operations agent that feeds stock levels.

discit Archive

A prototype "AI ERP": an LLM sales agent qualifies leads and hands orders to an operations agent.

Hackathons

Short builds for hackathons and competitions, made with AI coding agents that I directed.

Holdpoint In progress

A concrete crack-inspection tool: it finds cracks in a photo, gives their width as a millimetre band using a scale card, and leaves the decision to a human inspector.

Archwright In progress

A mixed-reality puzzle for Meta Quest: build a stone arch on your desk, and a masonry statics solver decides whether it stands. Playable in the browser; not yet tried on a headset.

PourProof In progress

A signed digital passport for concrete pours: simulated sensor readings checked by a smart contract on a local test chain, a strength range, and a carbon claim allowed once per pour.

Illustration of a family care log: a day of medication doses as pills on three timelines between a dawn and a moon mark, with a faint copy of the midday dose lifted above its lane, ringed in dashed ochre with a pause badge and marked 'duplicate · ask first' instead of being logged.

Baton In progress

An MCP server for family care logs that asks before logging a duplicate dose and leaves medication names out of replies to screenless clients.