About Collinear AI

Models learn in the worlds we build.

Collinear AI builds the realistic RL environments, verifiers, and training data that the largest frontier labs trust for their hardest model improvement problems.

Explore our work
Join the team
A large group of Collinear team members in matching black 'Harder Gyms' t-shirts, fist-bumping in an office lounge with large windows.
Team members working at shared desks in an open office, with laptops and coffee mugs, one person standing at a height-adjustable desk.
The Collinear team standing together in matching 'Harder Gyms by Collinear AI' t-shirts in front of large windows and potted plants.
A group of team members in matching t-shirts walking together down an office hallway.
A team member working at a standing desk, back turned to the camera, wearing a 'Stronger AI is born in harder gyms' t-shirt.
A canvas tote bag printed with 'Harder Gyms by Collinear.ai' and a stone-statue illustration, resting on a desk next to a water bottle.
Three team members standing with arms crossed, backs to the camera, wearing matching 'Stronger AI is born in harder gyms' t-shirts.
Why Collinear

Hard tasks.
Fair signal.
AI scale.

As models improve, the easy gains disappear. Frontier teams need hard, realistic tasks and trustworthy feedback, delivered in the current training cycle. Solving the challenge of delivering quality data at AI speed is why we created Collinear.

The Collinear team gathered in the Sunnyvale office for an all hands, people at their desks turned toward colleagues presenting beneath the Collinear sign.
Research-led

Research First.

We’re researcher-founded, researcher-led, and staffed by researchers. Our work on long-horizon agents, simulated users, reward design, verifier reliability, and benchmark design is shaping the next generation of AI.

Read our research
Four Collinear researchers seated together in the office, in front of the lit Collinear sign.
Our founders

An AI company built by researchers.

Collinear AI was founded by Nazneen and Soumyadeep to solve the data challenges holding back frontier-model improvement.

Nazneen Rajani

Nazneen Rajani

Co-founder and CEO

Nazneen’s work spans model evaluation, safety, alignment, and post-training. Before founding Collinear AI, she led post-training research at Hugging Face and Salesforce Research.

Soumyadeep Bakshi

Soumyadeep Bakshi

Co-founder

Soumyadeep’s work spans data strategy, verification, and evaluation infrastructure. Before Collinear AI, he led Deloitte’s AI practice, advising frontier labs on data for model development.

Our investors

Investors who get the frontier.

Engineering Capital
112 Capital
B Capital
Storm Ventures
Hitachi Ventures
Join Collinear

Grow fast with Collinear.

We’re growing the team every week to meet demand from the frontier. If you want to work at the heart of AI and join early enough to shape the company, we’re opening a dedicated San Francisco office alongside our Sunnyvale headquarters.

A large group of Collinear team members in matching black 'Harder Gyms' t-shirts, fist-bumping in an office lounge with large windows.

How we work

We work in hours and days, not weeks and quarters. Researchers and engineers own the industry’s biggest research problems end to end.

See open roles
Merch

We’ve got swag.

Beanies, tees, water bottles, stickers — made for the team, happy to share. We’re not set up as a shop yet, so if you’d like something, just ask.

Email us about swag

Company and press

Collinear AI builds realistic RL environments, verifiers, and training data for frontier models. Trusted by leading frontier labs, Collinear AI is headquartered in Sunnyvale, California, and is opening a San Francisco office.

CWE-BenchPress inquiriesContact us
Orange twisted ribbon-like shape forming a loop on a black background.
Get started in minutes

Better simulations.
Better data. Better agents.

See what a thousand rollouts can teach your agent in 30 minutes.

Talk to a Researcher