Member of Technical Staff, Environment Engineering

Ambitus works with enterprises and frontier AI labs to improve how AI performs in professional settings, starting with engineering and life sciences. The mission is to automate white-collar work at firms where the gains show up in daily life, in physical engineering and life sciences, more than in purely digital tools. Customers are mid-market enterprises doing more than $50 million in revenue, and frontier labs. Ambitus is already working with three of the leading labs. People from Ambitus sit inside enterprises and automate workflows. On the back end, an automated eval platform builds evals for those custom automations, which shows where models fail and feeds the reinforcement-learning environments Ambitus sells to the labs. The loop is to embed with a company, find the failures, sell the data that improves the models, then expand inside that enterprise.

Ambitus raised a $4 million seed from Neo, with angels from OpenAI, Thinking Machines, and Summit Partners. It is at an eight-figure run rate and growing fast. The team is 6 people, going to 8, with an end-of-year goal around 15. A co-founder was early at Mercor, and the founding group includes Stanford and BCG backgrounds. Seed stage, about 1 to 10 people, founded in 2026, in San Francisco.

Ambitus is hiring a member of technical staff for environment engineering. You build the systems that turn real professional work into tasks models learn from, and into evaluations researchers use to measure progress. A typical input is a business that has CAD drawings and the matching 3D model. You build scalable pipelines that ingest that data and produce usable reinforcement-learning tasks or evals, either to evaluate Ambitus's own agents or to sell to a frontier lab. You define what a dataset should test with lab researchers, review initial samples before production scales, develop quality checks and grading methods with domain experts, catch ambiguous instructions and reward-hacking tasks, and investigate failures so the next dataset is better.

There is no single exact skill they screen for, because a lot of this is new if you have not worked at a data company. What matters is being genuinely technical and a strong problem solver. Two shapes fit. Someone with real data-company experience, such as Surge or Mercor, who has done similar work and can start faster. Or a former or current quant who wants more real-world impact. The data-company background is the stronger fit, because they understand the technical challenge faster. The useful pickup from a large data company is core data delivery or environments. Payments or another non-core team does not carry the know-how. Data companies have grown a lot, and not every engineer from one is relevant.

You own the pipelines that turn enterprise data into reinforcement-learning tasks and evals. You work with lab research teams on dataset requirements and sample review. You own quality checks and grading with domain experts. You investigate review and delivery issues and feed the findings into later datasets.

Bar: about 1 to 4 years. Strong programming, and experience building data pipelines, automation, or AI applications. You turn a research question into a concrete technical approach and judge whether it works. You inspect individual examples and chase inconsistencies. You communicate clearly with researchers and subject-matter experts while requirements are still forming. Model evaluation, synthetic data, or agent systems are bonuses. Core data-delivery or environments work, a quant background, side projects you shipped on your own time, and staying calm through a competitive data business are green flags. A data-company resume on non-core work, dwelling on losses, needing fully scoped predictable work, or an ego that clashes with a small team is a miss. Brand-name signal helps.

In office in San Francisco. Visa sponsorship is limited. Equity is about 0.1% to 0.5%. A free gym membership, health, dental, and vision, and a $1,000 a month meals stipend. Posted work intensity is 9-9-6. The process is a one-hour whiteboard system design, a one-hour open-ended coding problem where AI agents are allowed, a one-day on-site on an open-ended project taken as far as you can plus a final presentation, then an offer.