Member of Technical Staff - Post Training

Warp Sciences; San Francisco (CA); On-site; USD 180,000 - 280,000; Full time

Job summary

Warp Sciences is building scientific infrastructure in the United States to accelerate discovery of new medicines. We bring AI, robotics, and life sciences together in increasingly automated labs designed to run experiments, learn from the results, and improve with every cycle.

The post-training research engineer or scientist will own the end-to-end post-training loop, improving RL capabilities, evaluation systems, and deployment into real products across Warp's robotics and foundation-model

Qualifications

• Track record of meaningful results with systems that support rapid experimentation.
• Shipped improvements into real products and learned from deployment.
• Strong hands-on experience with reinforcement learning.
• Eval-led approach: define success before optimization and verify gains.
• Data-driven approach: identify, create, and select data that yields measurable improvements.
• Helpful: robotics, embodied AI, or learning from interaction.
• Helpful: post-training multimodal, perception, or action-generating models.
• Helpful: balancing model quality with latency, compute, and deployment constraints.

Responsibilities

• Own the post-training loop from capability gap through evaluation, data development, training, deployment, and production feedback.
• Build evals grounded in product needs with clear baselines and held-out tests.
• Create experimentation systems for fast, reproducible hill climbing and easy result comparison.
• Develop and improve RL training methods, reward signals, rollout generation, and supporting infrastructure.
• Curate, generate, and refine training data based on model failures, evaluation results, and product feedback.
• Improve model behavior across Warp’s robotics and foundation-model stack in collaboration with robotics, infrastructure, and product teams.
• Validate gains through ablations, robust comparisons, and testing beyond the training distribution.
• Ship improvements, measure production impact, and inform next iteration.

Skills

Reinforcement learning

Evaluation design

Experimentation systems

Post-training research

Product-focused ownership

Job description

Warp Sciences is building scientific infrastructure in the United States to accelerate the discovery of new medicines. We bring AI, robotics, and life sciences together in increasingly automated labs designed to run experiments, learn from the results, and improve with every cycle.

Warp does not have a proprietary drug pipeline. We build the infrastructure that enables research teams to discover new medicines faster.

Warp is looking for a post-training research engineer or scientist who has worked at the frontier and knows how to turn strong models into better products.

Our AI stack spans local robotics and perception models, cloud foundation models, and the runtime connecting them. You’ll own post-training work that improves these systems’ capabilities, reliability, and performance in real use.

The most important qualification is a track record of meaningful results: you’ve built systems that support rapid, rigorous experimentation, achieved convincing state-of-the-art improvements, and shipped those improvements into real products. You lead with evals, treat data as a core part of the work, and bring deep hands-on experience with reinforcement learning.

What you’ll do

• Own the post-training loop from identifying a capability gap through evaluation, data development, training, deployment, and production feedback.
• Build evals grounded in product needs, with clear baselines and held-out tests that distinguish real progress from benchmark overfitting.
• Build experimentation systems that enable fast, reproducible hill climbing and make results easy to compare and diagnose.
• Develop and improve RL training methods, reward signals, rollout generation, and supporting infrastructure.
• Curate, generate, and refine training data based on model failures, evaluation results, and product feedback.
• Improve model behavior across Warp’s robotics and foundation-model stack, working closely with robotics, infrastructure, and product teams.
• Validate gains through ablations, robust comparisons, and testing beyond the training distribution.
• Ship improvements, measure their impact in production, and use what you learn to drive the next iteration.

What we’re looking for

• Experience working at the frontier of model capabilities, with meaningful personal ownership of post-training results.
• A demonstrated ability to achieve state-of-the-art results and substantiate them with rigorous evaluation against strong baselines.
• Deep practical experience with reinforcement learning, including reward design, rollout systems, training stability, and diagnosing failure modes.
• An eval-led approach: you define what success means before optimizing and scrutinize whether measured gains reflect the intended behavior.
• Strong data intuition, backed by experience identifying, creating, and selecting data that produces measurable improvements.
• Strong engineering skills and experience building reliable training and experimentation systems.
• A track record of shipping model improvements into real products and learning from deployment.
• An outcome-driven mindset: you follow the evidence, move quickly, and take responsibility for whether your work improves the product.

Helpful, but not required

• Experience with robotics, embodied AI, or learning from interaction.
• Experience post-training multimodal, perception, or action-generating models.
• Experience balancing model quality with latency, compute, and deployment constraints across local and cloud environments.