Soumi De
I like solving hard things.
Dream. Believe. Engineer.
Human data, before it was sexy.
Quoted in
I was in human data before it was sexy: first at Mercor, back when it had no revenue, and now at Handshake.
Most of that time has gone into starting things. I keep ending up at the beginning: a new line of business, a team that doesn’t exist yet. I get it working, then either scale it or hand it to someone who’ll run it well.
Today I’m Head of Strategy and Operations at Handshake. Before that I was Head of Growth at Mercor, where I’d joined as the first operations hire.
Built and scaled
Multilingual
Scaling$0 → $100M ARROne quarter31 languagesOn track for $250M
I took multilingual from zero to $100 million in annual recurring revenue in a single quarter. Today we’re one of the most reliable vendors across the 31 languages we support globally, with a path to $250 million that we’re already on track for.
It was a solo sprint. I had to understand what frontier models could and couldn’t do in each language, hire the people, and build the business as separate verticals, one per language and locale. Then the supply growth, retention, infrastructure and product roadmap for every one of them. I’ll soon hand it to a permanent owner.
AI Safety, International
Scaling2x revenue goalNew global marketsFull-time team from scratch
I’m building the team that will double our red teaming and safety revenue.
I own our expansion beyond the United States, and we’ve already entered several global markets, safely. I set up dedicated red teaming teams and capabilities, plus wellness support so the fellows and candidates doing this demanding work are looked after. Now I’m hiring a full-time team from scratch, alongside contingent experts, to give it all a lasting structure.
STEM and Code
Handed offI started both verticals, got them running, and handed them over.
Product and operations
Operator platform roadmapping
LedAnalysis and quality checks are engineering problems, not vibes. I roadmap the operator platform around that idea, so the checks that matter get built into the product and run the same way every time, instead of living in someone’s judgment.
ProductOps initiatives
LedI don’t like solving the same problem twice. Across several ProductOps initiatives, I took the operational problems I was fixing by hand and turned them into product roadmaps with engineering, product and design, so the fix lives in the product instead of in someone’s calendar.
Internationalization
LedLed internationalization across several products, on both sides of the marketplace: the experts doing the work and the companies who need them.
Expert Network growth
LedWorked out where experts come from, which channels were worth it, and how each new person could bring the next one.
Hiring
AlwaysThe top priority on everything above.
I quit class. Not the exams.
In my sophomore year, Mercor asked me to join full time. It had no revenue yet, and I’d be its first operations hire. I said yes, which in practice meant quitting university for everything except exams.
Why leave? Fitting in is easy, and falling into the box is even easier: class, degree, job, one tidy step after another. It’s much harder to run at full speed toward something other people can’t see yet.
I could see myself spending university getting ready for a job while the problems I cared about were already out there. I didn’t want to wait four years to be allowed near them.
Did it work? I worked full time, came back for exams, and still finished near the top of my class. Exams turned out to be one more hard problem, and by then I’d had a lot of practice with those.
Hundreds of hackathons. For fun.
Rewind a year. In my first year of university, I entered hundreds of them and won a lot. The prize money went straight into software I hadn’t tried yet. Then I signed up for the next one.
Build → Win → Buy new software → Build better ↺
I was studying Computer Science at the time. Those hackathons are also how I ended up working full time at Mercor a year later.
Quiet is not the same as alone.
Around twelve, I’d sneak out onto the road with a camera and photograph whatever was going on.
One winter I spotted an old woman alone on her balcony. Quiet, still, nobody around her. I decided she was lonely, and probably poor. I raised the camera.
She saw me, and she was furious. Then five people from the neighbourhood started chasing me.
She wasn’t alone. She was quiet. The people around her were invisible from the street, right up until they came running.
Humans are humans are humans.
The people are the work.
Relationships matter more than anything else I do, by a long way.
Thinking about every situation like an empath, from inside the other person’s shoes, has gotten me most of the places I’ve been. The moment you lose that and start optimizing for your number, you lose the people who were going to help you win.
I’ve valued people more than anything else in my life, and I would do it all over again. There’s no algorithm for them, and nobody is replaceable.
People mistake entropy for work done.
The quietest work is usually the work holding everything up.
I read that balcony wrong because it was quiet. I see the same mistake at work all the time. In human data, the loudest project often looks the most alive: constant pings, busy dashboards, lots of motion. Quiet gets read as empty, and noise gets read as progress.
Meanwhile the experts and operators quietly holding everything together are easy to overlook, until something breaks and you find out how many people were there all along. I try to notice them before that.
1,200 films and shows. One perfect biryani.
Films that make me feel something.
I’ve watched over 1,200 movies and TV series. I know the number because I rate every one of them for a friend. Any feeling counts, including the ones I’d rather skip. A few from the log:
A Clockwork Orange
1971Felt Uneasy
The Silence of the Lambs
1991Felt Breath held
The Departed
2006Felt Trust nobody
Requiem for a Dream
2000Felt Wrecked
Food that tastes like home.
I find almost every kind of food interesting. These are the ones that taste like home.
Kolkata & Mom’s kitchen
Order 1,201 · Table for one
- Kolkata mutton biryaniwith the potato
- ChaanpKolkata’s own
- Mom’s pulaonothing comes close
- Mom’s kosha mangshostill the benchmark
- Diet Cokeon the side
Thank you, come again
Right now, a lot at once.
At work, I’m scaling several new data verticals, running evals, and helping shape products that make the work easier.
Outside of it, I’m finding the next film to rate, over a plate of biryani and a Diet Coke.
If you’re starting something, or you’d like to argue about The Departed, I’d love to hear from you.
Email soumizde@gmail.com LinkedIn /in/soumizde ↗ X @soumizde ↗