saim shafique
ai systems engineer with a verifier’s eye
about
tldr; i’m saim shafique (sx4im), an ai systems engineer in hafizabad, pakistan. i build ai tools, mostly the kind that check other ai tools. right now my time goes to two things: contributing to open source, and building bmo: my main project, and the thing everything else fits around.
just a boy with a linux and an ide: i write eval data, prompt agents to do real engineering, and build small open-source tools that solve one problem well. :)
honestly, i’m a bit annoying about correctness. if an agent says it ran the tests, i want the receipt. if a test only fails sometimes, i’ll replay it a hundred times until it confesses. “looks fine” is not a standard i can ship on, and i think that comes from breaking enough software to know how quietly it lies.
i also like being there before anything gets trusted: figuring out what “correct” even means for a task before a model takes a swing at it. and if something can’t be verified, i’ll say so early. i’d rather ask an awkward question on day one than rubber-stamp the wrong thing on launch day.
right now
- building
- bmo: my main project, and where most of my building time goes right now.
- contributing
- trending repos: the busy, fast-moving ones everyone watches.
- interested in
- small teams and startups that verify before they ship. also into cyber, and gardening when i log off.
github activity
@sx4im · nov 2025 to oct 2026experience
made benchmark tests for datacurve’s deepswe, on the olympus and terminus projects, and earned $1,100 from the work. tests built to separate agents that genuinely solve problems from agents that only sound like they did.
i still think that made me a stricter engineer. once you’ve written the trap, you stop falling for confident-sounding output.
author of resultseal, skillcheck, chronos and more: open-source tools and 2,000+ contributions, all aimed at one thing: engineering you can verify. the full list is down in oss .
projects
bmo
websitethe main project i’m building, and where most of my time and attention goes. everything else on this page is what happens around it: open-source tools, merged prs in other people’s repos, and the odd agent broken on purpose.
resultseal
githubobservation verification engine for agentic tool use. every tool observation is sealed and verified before it’s trusted: if it can’t be proven, it doesn’t pass. no exceptions, no “looks fine.”
chronos
githubdeterministic simulation testing for node.js. pins time, timers and scheduling to a seed, so unpredictable async behaviour replays the exact same way every run, and a flaky failure becomes one you can actually fix.
skillcheck
githuba/b testing cli for ai agent skills. run two versions of a skill on the same tasks and see which one truly performs better: same tasks, same scoring, one winner. no vibes, no cherry-picked demos.
oss.
my contributions: 72 merged prs across 16 external repos. i ship my own tools on weekends, and fix other people’s codebases in between. every row links to the actual merged pull requests.
tech stack
( hover to play )- typescript
- javascript
- react
- next.js
- tailwind css
- motion
- node.js
- python
- flask
- postgres
- prisma
- supabase
- mongodb
- redis
- firebase
- playwright
- docker
- linux
- git
- github
- github actions
- vercel
- cloudflare
- figma
say hi
i’m always up for a chat about ai agents, evals, or something completely unrelated. you know where to find me.