01
Ship real things
A demo that never reaches users is a very expensive screensaver. Every engagement we take is scoped to end in production — with real traffic, real users, and a real number attached.
The studio behind the systems
GENTRIC.ai was founded by ML engineers who were tired of AI theater — the decks, the demos, the pilots that never ship. We started a studio where "done" means running in production, with evals to prove it.
Our story
Built by engineers who got tired of waiting
GENTRIC.ai was founded in 2021 in San Francisco by a small group of ex-FAANG ML engineers who kept watching the same movie: brilliant models trapped in notebooks, six-figure consulting decks gathering dust, and "AI initiatives" that never met a real user. We left to build the firm we wished our employers had hired.
Our founding thesis hasn't changed since day one: the gap isn't models, it's engineering. The models are astonishing and getting cheaper by the quarter. What's scarce is the discipline to wrap them in data pipelines, evaluation harnesses, and operations that survive contact with production traffic. That's the part we do.
We bootstrapped from our first client check and have been profitable every year since — no investors, no growth-at-all-costs, no pressure to sell you anything you don't need. Today we're a 30-person, senior-only team across San Francisco, London, and Bengaluru, and we still turn down more projects than we take.
2021
Founded in San Francisco
Four engineers, one sublet office on Folsom Street, and a first client — a logistics firm whose forecasting model we took from notebook to production in nine weeks.
2022
Gentric Core v1 ships
We extracted the eval harnesses, retrieval pipelines, and observability we kept rebuilding into one accelerator platform. Every engagement since has shipped on it.
2023
Tenth production system, London office
Our tenth system went live and our first European clients pulled us across the Atlantic. A three-person London team opened in Shoreditch that autumn.
2024
SOC 2 Type II, 40th deployment
We earned SOC 2 Type II certification and crossed 40 production deployments — including our first regulated-industry systems in health and finance.
2026
250M+ daily predictions, Bengaluru hub
Client systems now serve over 250 million predictions a day. Our new Bengaluru engineering hub keeps the sun from ever setting on an on-call rotation.
What we believe
Four rules we don't bend
Every studio has values on a wall somewhere. Ours are enforced in contracts, code review, and who we hire.
02
Evidence over hype
We talk in eval scores, latency budgets, and unit economics — not vibes. If we can't measure that a system works, we don't call it done, and we won't put it in a case study.
03
Your IP is yours
Everything we build in your engagement — models, pipelines, prompts, weights — belongs to you, in your cloud, from day one. We keep the scars and the lessons; you keep the assets.
04
Senior talent only
No bait-and-switch staffing. The engineers you meet in the sales call are the engineers who write your code. Median experience on our team is eleven years, and there is no bench of juniors behind them.
Leadership
The people who answer for the work
Six leads, zero layers between them and the code. Every one of them still ships.
AM
Arjun Mehta
Co-founder & CEO
Ex-Google Brain infrastructure lead. Wrote GENTRIC's first invoice and its first eval harness in the same week — still reviews both.
EV
Elena Vasquez
Co-founder & CTO
Former Meta ML platform architect. Designed Gentric Core and has personally load-tested every release since v1.
TN
Tomasz Nowak
VP Engineering
Runs delivery across all three offices from London. Known for killing scope creep in one meeting and shipping a week early anyway.
PR
Priya Raghavan
Head of AI Research
Published on retrieval and evaluation before it was fashionable. Turns papers into production techniques about six months ahead of the market.
MC
Marcus Chen
Head of Data Platforms
Built streaming infrastructure at Stripe. Believes most "AI problems" are data problems wearing a disguise — and is usually right.
AD
Amara Diallo
Head of Design
Leads product design from Bengaluru. Insists that an AI system users don't trust is a system that doesn't work, whatever the eval score says.
0senior engineers, in-house
0global offices — SF, London, Bengaluru
0junior handoffs, ever
0senior-led delivery on every project