## The company
Ooak Data turns company data into training data for AI agents.
Frontier labs can train models to reason. They cannot train them to work: navigating a real company's Slack threads, half-finished Notion docs, contradictory Jira tickets, and permission boundaries. That requires real enterprise data, and you cannot synthesize it. You have to source it.
We plug into enterprise tools, anonymize everything into a structurally identical digital twin, and generate reinforcement-learning environments with expert-level tasks calibrated against frontier models.
We are a Y Combinator company with 7 figures signed contracts with three frontier AI labs, and we are scaling delivery aggressively over the next twelve months. Three founders, full-time since December 2025: Pierre-Louis (CEO, ex-COO in edtech, sold data to frontier labs), Grégoire (CPO, first PM & head of Ops at Epsor through Series B), Thomas (CTO, ex-Head of Data at PayLead, ex-Samsung AI lab).
## Why this role exists
Our engineering team ships faster than founders can specify. We have four product surfaces, most of them internal, all of them technical, and every one of them decides whether an environment ships this week or next.
We are hiring our first Product Manager to own the product: what gets built, in what order, and whether it actually made us faster.
## What you will own
The full chain, from the moment a CEO agrees to share their data to the moment a frontier lab receives an environment.
**Partner experience**
* The Data Hub, onboarding and export tutorials a CEO sees when they hand us their company's history. Trust and speed at the top of the funnel.
**The data pipeline**
* Connectors, ingestion and anonymisation. How raw Slack, Drive, mail and CRM exports become a clean digital twin. You will spend a lot of time with Thomas (CTO) here.
**Pharos, our delivery platform**
* The internal tool our Ops team lives in to run QA and delivery. Throughput is the metric. This is where 3 env a week becomes 10.
**RL environments**
* The thing the labs buy. Task design, calibration, what separates a dataset from a real agent training ground.
You will run two kinds of work at once: short projects that need tight coordination across Tech, Ops and GTM this week, and long ones where nothing exists yet and you get to decide what should.
## Who you are
* **A doer.** You know the product methodologies and you use them when they help, not by the book. Your instinct when something is slow is to fix it, not to schedule a workshop about it.
* **AI-pilled.** You vibe code. You prototype before you write a spec. You design your own mockups. You are curious about how the models themselves are built and trained.
* **UX taste, ROI brain.** Our workflows have to be brutally efficient. You judge a piece of software or a workflow by efficient it allows us to be or what value it brings to our customers.
* **You keep pace.** Our engineers ship a lot. You turn that into something coherent rather than slowing it down.
* **7+ years in product**, at least some of it technical: data, infra, ML, developer tools or ops platforms. Founders and first PMs who built the function from nothing are very welcome.
* **Fluent French and English.** Our customers are American, our data partners spread worldwide, our team speaks both.
* **Based in Paris.**
**The extra that make the difference**
* A research streak: you read papers, you have opinions on RL, you have trained or fine-tuned something.
## Terms
* **Compensation:** €70-90k + bonus, and 0.2% to 0.5% BSPCE depending on profile.
* **Location:** Paris Morning Laffitte, on-site with 1 to 2 days WFH.
* **Start date:** as soon as you can.
* **Perks:** Alan health insurance, 50% Navigo.
Visa sponsorship is offered for this role.
