About · Arjun Moudgil
Big-league rigor, corner-shop manners.
Five years on Meta's AI Infrastructure · Twelve in the industry · NYU
01 — The infrastructure years
For five years, Arjun Moudgil worked on the performance, reliability, and efficiency of Meta's AI Infrastructure — the large-scale training clusters of tens of thousands of GPUs behind models like LLaMA 3. Keeping machines at that scale fast, reliable, and easily debuggable — improving the observability by magnitudes, year after year.
Engineering at that scale is a strange discipline: a wasted hour costs more than most small businesses gross in a year, so nothing is allowed to stay mysterious. Every failure has a cause. Every cause can be found.
The Meta years also spanned the apps themselves — reliability work on Instagram, WhatsApp, and Facebook for iOS, keeping them dependable on a billion phones. Before Meta, the same craft at Turo — and underneath it all, a Master's in Computer Science and AI from NYU.
The discipline built for planet-scale AI,
pointed at a small business's Tuesday.
02 — The turn
The turn came from noticing who that rigor had never been offered to — and where it would matter most. Not another platform for companies that already have ten. The businesses he actually lived among in New York City: the accounting practice with four staff, the law office, the shop with the handwritten ledger. Small business employs nearly half the country, and it's exactly where software and AI can do the most good per dollar — yet for decades, custom-built meant for big companies only, because bespoke software cost too much to make for anyone else.
AI changed that arithmetic. Made-to-measure is finally affordable at corner-business scale — but somebody has to be the one who hands it down. That's the job.
03 — The practice
How he works, in one paragraph: watch first, build second. Smallest fix first, so trust is earned on results, not promises. No hostage situations: the client's data and money accounts in their name, the code theirs on request, everything documented in plain English. Fixed prices, quoted before work starts. Delivery is never the finish line: the maintenance and support that follow are held a notch above the build itself — whatever heart went into the making, the care after it aims higher. And when the honest advice is buy nothing, that's the advice.
Watch first, build second
The audit precedes every build. No exceptions.
Smallest fix first
Trust is earned on results, not promises.
No hostage situations
Your data, your money accounts, and the code on request — documented in plain English.
Care, a notch above
After delivery, maintenance and support are held a notch higher than the build itself.
Come tell us about your Tuesday.
The audit starts with watching how the work actually happens.