Two years in AI human data. Built Mercor’s Talent Success vertical from scratch as the company went from $1B to $10B. Now Operations Lead on the founding team of an early-stage AI human-data startup.
I run the operations that turn expert human judgment into training data for AI labs. Evenings and weekends, I ship iOS apps.
Path
Engineer first, then two years at the sharp end of AI human data.
2026 to now
Operations Lead, remote
Founding member running operations end to end for an expert human-data platform: strategic projects, talent success, compliance, payments, client and vendor relationships, tooling. Built the onboarding system from scratch and cut domain-expert onboarding time by about 80%. Lead a distributed, cross-functional ops team.
Startup ops · RLHF · Team leadership
2024 to 2025
Associate to Senior Success Engineer, remote, Sep 2024 to Nov 2025
Built the Talent Success vertical from scratch as the company went from $1B to $10B: expert onboarding, AI-lab delivery, quality and retention, then owning the lab relationships. Repeatable processes for onboarding and client delivery through a period of explosive demand.
Talent Success · AI-lab delivery · Hypergrowth
2021 to 2024
Senior System Engineer, Bengaluru, Sep 2021 to May 2024
Software development across enterprise systems. Built and maintained production code. The engineering foundation I now apply to operations: automate the repetitive, instrument everything, scale what works.
Software development · Enterprise systems
Apps
TumeraMusic taste (opens in a new tab)
HadoraFive-tone replies (opens in a new tab)
ScreenAwayTime offline (opens in a new tab)
NocteraAI ad-block browser (opens in a new tab)
DynaProDynamic Island pet (opens in a new tab)
ClearwingTransparent widgets (opens in a new tab)
DroploOne-tap water (opens in a new tab)
SpilloVoice journal (opens in a new tab)
UmmoSpeaking coach (opens in a new tab)
TypekataType to fight (opens in a new tab)
DotwallDot wallpapers (opens in a new tab)Notes
Judgment
The frontier race is leading somewhere strange. We’re building models that can answer almost anything, while getting worse at agreeing on what a good answer is.
Capability is moving fast. Judgment is the bottleneck. That gap is where the next chapter of AI will be decided.
Operations
Model quality stopped being a GPU problem. It’s a people problem now.
I’ve seen great reviewers improve data quality more than any pipeline fix. The next AI moat isn’t architecture. It’s operations.
Benchmarks
Every AI benchmark has a hidden step: a human deciding what “correct” means.
Same model. Same test. Different humans, different score. So when a company says “our model is #1 on benchmarks,” I ask who wrote the grading rules?
Instructions
AI companies keep saying they need more data. Most of them don’t. They need better instructions.
Add 10,000 more tasks to that and you haven’t built a dataset. You’ve built 10,000 disagreements. Scale makes a good process better. It makes a vague one worse, faster.
