currently: 3 hackathon submissions in judging

tobi david obembe

builder · founder — abuja, running an agent fleet

0.22x → 1.31x

kernel speedup on muse spark across 21 measured iterations. public leaderboard, so you can check it yourself.

short on time? check these three

proof / open kernel challenge

kernloop is my gpu kernel pipeline from the open kernel challenge (meta x paradigm). it starts slower than the baseline and ends 1.31x faster, and every step in between was measured and logged.

whatvaluecheck it
starting point0.22xleaderboard
final result1.31xleaderboard
measured iterations21build log on x
target hardware / modelh200 / muse sparkthe challenge

between iteration 1 and 21 the pipeline went from failing compiles to sourcing every step by measured price and latency. promoted iterations landed 22% to 40% mean improvements; every failed iteration is logged with its reason. the failed rows are the point: the record is honest, not curated.

builds

products and businesses (ongoing, not experiments)

writing

shipped recently

an agent writes this list from the vault. it updates as things ship.

about

i'm tobi, a builder and founder in abuja. i started in crypto, went looking for better tools, and the tools became the products. today i run a fleet of ai agents that ships software, enters hackathons and keeps this page current.

right now i'm building hoolie (group buying for nigeria), running an agent-operated product portfolio, and writing gpu-brain, a public series that teaches ai hardware through the human brain.

if you landed here from an application or a hackathon: everything on this page links to something you can check. if you want to work together, the fastest route is a dm on x or the email below.