Ahmet Baris
Gunaydin
Full-stack developer
and researcher.
Consulting at Publicis Media (via bi:procsi / Azendo) on the dotstudio platform. Shipping research in the off-hours.
I look for the overweight step in a pipeline, ship a leaner replacement, and publish the measurement that produced the claim.
One question kept compounding.
markview was the first one — why is there no clean, secure way to share a markdown file beautifully? Then I looked at Chiang Mai's burning season and asked what would actually stop it. Tradition and convenience are the wall, not ignorance — so zero-burn turned rice harvest waste into biodegradable packaging instead.
Researching natural cleanup pulled me into mushrooms. Can we engineer better ones? Found AlphaFold. Then the obvious next question: why can't I simulate this on my Mac? Testing fitness functions on the GPU surfaced something obvious in hindsight — most of what GPUs do isn't compute, it's dispatch overhead. That became the kernel-fusion line.
Small-scale proof in. Time to push: Phi-3 in a browser without a compiler, the same forward pass visualised live, Geant4-DNA Monte Carlo ported to WebGPU, quantum chemistry in a browser tab. Because why not. The work, the commits, the open questions — live at the chronicle ↗.
Every project has a live URL.
These cards are rendered at load from /sites.json, a byte copy of the shared
file that generates the project metadata for every site in this set. Nothing on this
section is typed by hand, so a card here cannot disagree with the same card elsewhere.
Two preprints are archived on Zenodo.
Each record links the source and the reproducer that produced its numbers. Both DOIs below are concept DOIs, so they resolve to the latest version rather than freezing a reader on an old PDF.
Single-Kernel Fusion for Sequential Fitness Evaluation via WebGPU Compute Shaders
Evolutionary-compute benchmark on laptop and datacenter GPUs. The same advantage is measured across CUDA, WebGPU, JAX/XLA and Triton, so the effect is not specific to one GPU API or one vendor.
read → doi:10.5281/zenodo.19344276Single-Kernel Fusion for Autoregressive Transformer Decoding via WebGPU Compute Shaders
Transformer decoding in a single compute-shader dispatch: every operation for every token is fused into one kernel. It applies the dispatch-overhead thesis of the first paper to LLM inference.
read →The problem picks the language.
The work on this page spans TypeScript, WGSL, Python, Rust, CUDA and shell. What I check is whether the benchmark runs, the tests pass, and the domain resolves.
Code, benchmark, paper, deployed domain.
Two preprints archived on Zenodo, several live sites, and the repositories behind them. Every result on this page has a deployed artifact you can open and a source tree you can read.
I own the whole surface.
DNS, CI, edge caching, preview deploys, SEO, analytics, observability, SSO and billing, wired across this stack of projects and in prior senior roles.
Results ship with the code that produced them.
Every gpubench benchmark runs in your browser on your hardware, and every submitted run is public. The preprints link the source that produced their numbers. The cross-vendor benchmark data is published as an open snapshot and does not yet carry a citable DOI.
Where I've been shipping.
Day-job progression: founding engineer, then senior consultant at a Big Six media holding group. The research line above runs alongside it, in evenings and weekends.
Senior Full-Stack Consultant · dotstudio, Publicis Media
Consulting on the dotstudio platform at Publicis Media — one of the Big Six global media / advertising holding groups. Placed via bi:procsi; employed by Azendo. Full-stack engineering across a multi-tenant ad-tech platform.
Founding Engineer · Balance Cash (balancecash.io ↗)
First engineering hire at a seed-stage treasury fintech in San Francisco. Built SmartSweeps — automated cash sweeps into U.S. Treasury-backed, AAAm money-market funds for modern finance teams and CFOs. SEC-regulated investment advisor, SOC-audited, SIPC-insured. Backend, frontend, infra, and the compliance surface that made all three audit-passable.
I take research collaborations and senior engineering conversations.
Currently consulting full-time at Publicis Media on dotstudio (via bi:procsi / Azendo). Open to research collaboration and to the right senior staff or principal conversation.