Member of Technical Staff, Environment Engineering
- Build pipelines that turn enterprise data into RL tasks/evals
- Former quants + core data-company engineers (Surge/Mercor core)
The environment engineer builds the systems that turn real professional work into tasks models learn from and evaluations researchers use to measure progress. Concretely: Ambitus goes into a business that has, say, CAD drawings (a blueprint) and the matching 3D model, and the environment engineer builds scalable pipelines that ingest that data and produce usable RL tasks or evals, either to evaluate Ambitus’s own agents or to sell to a frontier lab. The work includes defining what a dataset should test with lab researchers, reviewing initial samples before scaling production, developing quality checks and grading methods with domain experts, catching ambiguous instructions and reward-hacking tasks, and investigating failures to improve the next dataset.
Akshana’s read on the ideal profile: if you have not worked at a data company, much of this is new, so there is no single exact skill he screens for. What matters is being genuinely technical, higher-intelligence, and a strong problem solver. In practice two archetypes fit: someone with real data-company experience (Surge, Mercor, and similar), who hits the ground running faster because they have done very similar work, and former or current quants looking for more real-world impact. The data-background person is more ideal because they understand the technical challenge faster. One important nuance from his pipeline: at big data companies, a candidate who worked at Mercor or Scale but on something non-core (payments, say) does not carry the know-how; the valuable pickup is someone who worked on core data delivery or environments. Keep in mind that Data companies have grown a lot and not any engineer is going to be relevant here.
What You’ll Own
- Scalable pipelines that ingest enterprise data and produce usable RL tasks and evals
- Working with lab research teams to define dataset requirements and review samples before scaling
- Quality checks and grading methods with domain experts; catching ambiguous, incorrect, or reward-hacking tasks
- Investigating review and delivery issues, fixing causes, and feeding findings into later datasets
Requirements
- Strong programming; experience building data pipelines, automation, or AI applications
- Genuinely technical, high-intelligence, strong first-principles problem solver
- Can turn a research question into a concrete technical approach and judge whether it works
- Careful about data quality: willing to inspect individual examples and chase inconsistencies
- Clear communication with researchers and subject-matter experts when requirements are still forming
- Bonus: experience with model evaluation, synthetic data, or agent systems
Signal
- Real data-company experience on core work (data delivery, environments) is the strongest fit (e.g., Surge, Mercor core teams)
- Former/current quant (or quant internship) looking for real-world impact
- Brand-name signal on the resume helps (he values it and finds it snowballs with other legible candidates)
- Side projects and demonstrated agency (how they spend their free time; shipping on their own)
Who Will Thrive Here
- A former quant or data-company engineer who wants real-world impact and direct work with AI researchers
- A high-intelligence problem solver who is comfortable in a new, open-ended domain Someone meticulous about data quality who will inspect individual examples
- A confident “winner” who handles the roller coaster of a competitive data business without dwelling on losses
- High agency and incentive-motivated (gets excited by the chance to compete and earn commission)