Zaid Ali Syed · Aurangabad
I'm Zayd, nahi nahi, Zaed, matlab Zaid
I am Zaid.
Even my name needed a correction. Wispr Flow kept what I meant. So here is my life, told the same way. White is the road I took. Pink is every road I almost took, and every time I said nahi actually.
Scroll to walk it · drag to look around · click anywhere for the music · read along with me at every stop
Podar International School
through Grade 10
Every road starts somewhere. Mine starts at Podar International School, where I studied until grade ten.
MGM University
B.Tech IT, class of 2027
Then MGM University, B.Tech in Information Technology. I came for a degree, and then did everything possible to live college life.
The semester I almost left
and the friends who stayed
Then college turned on me. One semester was so hard that I went to ask for my transfer certificate. I was done. Then a few friends showed up when I needed them most. I never collected it. I stayed.
The IEEE student branch room
MGM University
The boy who almost left became the one who brought people in. At the IEEE branch I recruited 105 organizers and won the best recruiter award, and our events reached 600 to 700 people. In this room we decided every event, made fun of each other for hours, and became a family.
The ones I leave behind
my juniors · met in third year
In third year I met a few juniors who became everything. Here, each of them is a small flame on the college steps. Soon I walk on, and they stay. Of all the forks on this road, leaving them is the one I cannot correct.
The first correction
I started building in public, still believing:
Nobody merges a stranger's pull request.
Nahi actually, three maintainers did: tenacity, returns and gqlalchemy.
Neo-Kyoto
my GitHub profile, drawn in light · each room is one of my projects
Some of what I build already carries real people. Twenty-seven merged pull requests into Zabira Academy, a live learning platform I did not build. A support dashboard that agents and team heads use at Pratyaya Limited.
The second correction
I once told everyone, proudly:
The headline feature is live.
Nahi actually, I had never read the data source's terms. I cut the feature. Terms first, now.
Lotus Temple
Delhi · 27 marble petals, not one out of place
The third correction
Mujhe pakka yakeen tha. I ran 229 agent runs, completely sure:
They will prove that AI agents cheat on flaky tests.
Nahi actually, 0 of 60 runs did. My own instruction, "don't touch the tests", took correct fixes from 9 in 10 to 0. I published it with my wrong hypothesis left in.
Eiffel Tower
Paris, 1889 · allowed to stand twenty years · still standing
The fourth correction
A quality gate kept failing.
Pass the gate and move on.
Nahi actually, I stopped the project and left the gate red. A fake pass would have been a lie I kept paying for.
Sydney Opera House
Sydney, 1973 · every shell cut from the same sphere
The fifth correction
Which brings me to the correction I care about most. Six failing tests turned green, and I told myself:
Ship it.
Nahi actually, the AI had taught the test to click twice, and a visitor's first click was still being lost. I fixed the real thing, not the test. That gap, between something that looks impressive and something that genuinely works, is what I care about.
In my own words: "I spend a surprising amount of time building things that nobody has asked me to build. Sometimes it's an idea for an AI system, sometimes it's an experiment to see whether an existing system actually works the way people claim it does. I once noticed a coding agent making tests pass by hiding the bug instead of fixing it. I could have moved on, but I ended up building small apps with planted bugs and a test harness to investigate that behaviour. I think I care about the gap between something that looks impressive and something that genuinely works. I keep exploring that gap even when there's no deadline, grade, or reward attached to it."
The sixth sentence
Every sentence so far, I lived first and said later. This one I said first.
I predict the tip lifts unedited first messages by 9 points.
Locked in public on 31 October · the prediction, with its timestamp
Not tested yet. It is the first test I would run with you.
What I measured first, with my own voice: 5 of 14 of my Hinglish corrections came out the way I meant them, and my name was spelled right 12 of 17 times.
How it would run: people install Wispr Flow and dictate one real message. A random half get one setup tip; the other half get none. The next day each person says whether they sent the first transcript unchanged. To tell 9 points apart from chance, it needs about 500 people in each group.
The last fork
My name needed one correction. This road needed five. This fork needs none.
Wispr Flow. You are standing at this fork.
November part-time, full-time from mid-December 2026. Willing to relocate to Bangalore.