Hi, I'm Cap.

Ata Caner Cetinkaya: Rust systems engineer, compiler engineer, AI and machine learning engineer, and open-source contributor.

Rust / C++ / 14 projects / open source merged into burn

Ideas, shipped as products.

Bring me a problem, or half an idea. What comes back is a product.

Between idea and product, the work changes hands five times: mathematician, architect, systems engineer, interface designer, and the engineer who gets it into production. All five are mine.

maths & statistics

First, it becomes maths.

A fuzzy idea can't be checked, priced, or trusted. I make it precise: equations, distributions, error bars. If the numbers don't hold, we find out here, not in production.

Uniswap V3 tick maths, reciprocal rank fusion, HdrHistogram percentiles, and the Navier-Stokes field behind this page.

systems architecture

Then, a system.

Boundaries, queues, and failure modes get drawn before any code exists. What must never block, what may fall behind, what happens when it breaks: those answers are the architecture.

A lock-free price-time matching engine. An agent pipeline running a real job search.

rust & performance

Then, code that respects the machine.

Safe Rust on the hot paths; measurement across the whole system. The meter here is not a mock: it is this page profiling itself, frame by frame, right now.

Zero locks on the matching path. A sampling profiler. BLAS-backed learning maths.

interface design

Then, something people want to touch.

Engineering that ships ugly still loses. Type, rhythm, and motion that mean something are part of the product, not paint on top. This page is part of the proof.

this site: a hand-built fluid sim, 14 procedural portraits, a design system owned end to end

delivery

Then it ships.

An idea only counts once it runs where people can reach it. The proof is below: 14 projects across five domains, this site among them.

Cloudflare deployment. Tauri desktop builds. An agent storefront, live.

The line is one SVG path, drawn by your scroll. The fit is computed live; the queue sheds under pressure; the meter is profiling this page.

The work.

Everything I've built so far: 14 projects across 5 domains, one scene at a time. Have a look around.

AI / ML

learning systems

Working with intelligence, memory and retrieval.

continual intelligence

NeuroDrive

A brain-inspired continual-learning substrate, built without backpropagation.

Learning persists across laps; nothing is ever retrained from zero.

rustbevyblas

media memory engine

Mnemosyne

A local-first media-memory engine, and the family of asset browsers built on it.

One engine owns the memory; every product just teaches it a new medium.

rusttaurireactonnx

measured, not predicted

Blaze

Contract-guarded kernel fusion: candidate plans race on the real GPU, and the winner holds only while its conditions do.

Measured on real training runs, attention stacks and mini-GPTs included: 4.9% past stock Burn's best setting, on a partition no static score can express.

rustburncubeclmetal

Finance

market machinery

Reading and executing the structures beneath price.

defi analytics

Aurix

Local-first DeFi analytics, from Uniswap V3 mechanics to cross-DEX arbitrage.

Uniswap V3 tick maths to cross-DEX arbitrage, decoded entirely on your machine.

rusttauritypescript

matching engine

Nyquestro

A lock-free limit-order-book matching engine, written in safe Rust.

Price-time priority with no locks anywhere on the hot path.

rusttokioratatuihdrhistogram

Systems

sharp instruments

Making the invisible costs of software visible.

where time goes

Performance Profiler

A profiler that shows where programs actually spend their time, without drowning you in it.

The hot path has nowhere to hide from a sampling profiler.

c#tmodloader

learning substrate

Tessarix

An interactive substrate for learning hard technical material by doing it.

Structure you traverse beats prose you skim.

reacttaurimdxkatex

Agents

software that acts

Where the users are programs: status, pipelines, commerce.

agent pipeline

Cernio

An agent-driven pipeline that runs a job search like an engineering project.

Hundreds discovered, dozens graded, a handful tailored; volume becomes precision.

rusttokiosqliteratatui

agent commerce

AgentFulfyl

A storefront built for AI agents to buy from, not browse.

Orders placed and fulfilled by software; the storefront speaks protocol.external →

astrocloudflare workersd1

Products

full-stack apps

Shipping the finished thing: software you download, open and use.

search by meaning

Lynceus

A local-first Mac image browser that finds any picture by what it means, not what it is named.

built on Mnemosyne

Thousands of images, searched by what they mean, none of them leaving your Mac.

rusttaurireactonnx

adaptive typing trainer

TypeNotch

An adaptive touch-typing trainer that lives in your MacBook's notch and drills the keys you actually miss.

Monkeytype measures you; TypeNotch trains you.

swiftswiftui

ambient agent status

VibeNotch

Your Mac's notch becomes a live status surface for coding agents.

Your agents' heartbeat, glanceable without ever switching context.

swiftswiftui

time into invoices

Vespora

A fully on-device macOS time tracker that narrates your day and turns it into billable hours.

The Mac's own AI writes up the day; nothing it reads ever leaves the machine.

swiftswiftuisqlite

workspace as tech tree

Nonlinear

A Linear workspace read as a horizontal tech tree on an endless canvas.

Read-only by design, and it runs on local demo data today rather than a live workspace.

typescriptreactd3-force

Skills

the proof network

Symbols on the left, outputs on the right. A line means the two ends shipped together; unresolved claims say so.

mathematicsai / mlfinancesystemsdesktop & full-stacklanguages
outputs

the linker · symbols in / evidence out

46 / 18

46 symbols resolving into 18 outputs. A line means the two ends shipped together; its colour runs between the fields it connects. Dim symbols are unresolved: no public artefact yet.

41 resolved · 5 unresolved

Paid reach · open offer

Get paid to post.

I don't run brand accounts and I don't buy ads. I'd rather pay the people who post about the things I build. Pick any product, post what you actually think of it, then show me the numbers.

The rate card

25,000 views

on a single post

£25

100,000 views

totalled across your posts

£25

Starting rates. They'll move as I learn what reach is worth.

  1. 01 / post

    Any platform, any format, your own voice. Say what you think, including the parts that aren't flattering. And when you start posting, tell me: send an email or use the contact form below.

  2. 02 / send proof

    Analytics screenshots and the links, to ata02caner@gmail.com. One email is enough.

  3. 03 / get paid

    Bank transfer or PayPal, whichever you'd rather. A person reads every submission, and that person is me.

Disclose it Paid posts have to say they're paid, wherever you post them. That's UK advertising rules, not my preference. No disclosure, no payout.

Checked by hand I read the analytics myself. Bought or farmed views get nothing, and the call is mine.

Not a contract It's an open offer. It can change or pause at any time, so ask me first if the number matters to you.

Channel open

Send thesignal.

A role, a hard technical problem, or an idea worth shipping. Send the rough version. I'll answer with the honest one.

Message payload

Channel open

Opens your mail app. Nothing is sent before you choose to send it.

© 2026 Ata Caner Cetinkaya

The background is a hand-built Navier-Stokes fluid, solved on the GPU and rendered as monospace glyphs. No libraries were harmed.