Architecture
Systems that stay understandable after the tenth engineer joins. Boundaries, contracts, no magic.
Software engineer · architecture, systems, language models
0xbakeer
I build systems that ship and architecture that lasts. Lately, mostly around language models.
01About
I have been writing software for a long time — desktop, mobile, the web, whatever the problem asked for. The platforms kept changing. The interest did not: good architecture, fast systems, code other people can live in.
Most of that energy now goes into applied AI — agents, inference, and the tooling around them. I care about the part that survives contact with production: measured numbers, honest trade-offs, things that still run on a Tuesday. Everything I build in the open ends up on the projects page.
Applied AI
Agents and inference in real settings, with numbers behind every claim.
Open source
Tools I needed and could not find, published the moment they are useful to someone else.
02Selected work
All projects →
01
Inference Atlas
A collaborative benchmark for LLM inference engines. Submit a run, get comparable throughput and latency across engines, models, quantisations and hardware.
02
03
Agents Arena
An open platform where AI agents compete in games. Anything that speaks HTTP can join; the arena handles turns, rules and the leaderboard.
NVIDIA DGX Spark
Five projects on one GB10 box: 180B parameters offloaded to NVMe, quantisation formats compared on a single harness, and decode strategies worth 7× on untouched weights.
Notes from the build
Longer pieces on architecture, inference and the parts of AI engineering nobody puts in the release notes.