Member of Technical Staff - AI Training Data

Description

About Warp

At Warp Sciences, our mission is to help the world discover new medicines faster—and build the capacity to do it here in America. We bring AI, robotics, and life sciences together in automated labs that design, run, and learn from experiments, with each experiment making the next one better. We’re not developing drugs of our own. We’re building the infrastructure for the people who do.

Warp is building autonomous laboratories that can run scientific experiments end-to-end. Our work combines robotics, automation, and advanced AI to enable breakthrough discoveries in chemistry and beyond. High-quality real-world data powers our systems, models, and robots.

We are looking for a senior, hands-on AI data builder to own what our models learn from. You’ll set the data strategy and build the systems to execute it across our chemistry foundation model, our lab perception stack, and our robot-learning system. You will determine what data matters, where to source or generate it, and how to turn it into measurable capability gains. Your scope will span open datasets, synthetic data, RL rollouts, expert feedback, teleoperation, and physical lab execution. You'll connect these sources through Vault, our shared data storage and lineage system, making useful data faster to produce, easier to trust, and simpler to reuse as Warp scales.

Responsibilities

• Own the data agenda: Translate model failures and lab priorities into a collection and generation plan. Decide the right mix of open, licensed, synthetic, and real-world data. Prioritize coverage gaps, difficult cases, and highest-value expert involvement.
• Develop external data: Source and assess chemistry, scientific, and multimodal datasets. Evaluate licensing, provenance, quality, and relevance. Build extraction, normalization, and enrichment pipelines that convert promising sources into useful training and evaluation data.
• Build synthetic data & RL data loops: Generate examples, model rollouts, and simulated trajectories. Develop rewards, verifiers, and data filtering with researchers. Use evaluation and RL feedback to improve generators, sampling, and curricula. Test for reward hacking, contamination, and transferability of synthetic gains.
• Make physical experience useful for learning: Help design and conduct lab and teleoperation data collection with chemists and robotics teams. Capture demonstrations, corrections, failures, and recovery attempts, keeping all data aligned with the experimental context.
• Build AI-native curation workflows: Use models and agents to label, enrich, deduplicate, and mine data, with human review where it adds value. Own quality checks, dataset composition, and versioning. Make annotation and feedback efficient.
• Measure data by capabilities it creates: Partner with researchers on data ablation, supervised fine-tuning, and RL experiments. Use held-out evaluations and real failures to improve the data mix. Track capability lift, coverage, efficiency, and cost per useful example—not just dataset size.
• Make the shared data foundation work: Own the path from collection and generation. Build reliable ingestion, storage, lineage, retrieval, and dataset delivery with clean interfaces. Partner with infrastructure engineers to keep this process fast, dependable, and easy to maintain as we grow.

Requirements

• A track record of improving model capabilities through better data, with hands-on ownership from identifying a gap to building the dataset and measuring results.
• Practical experience with ML data development: synthetic generation, model rollouts, multimodal datasets, expert annotation, or real-world collection.
• Strong understanding of data composition, sampling, demonstrations, rewards, and verifiers, and how they affect fine-tuning and RL.
• Strong software engineering skills in Python, with experience building data pipelines, model-driven tools, and reproducible experiments.
• Good judgment on data quality, provenance, licensing, leakage, and evaluation, and the ability to inspect and critically analyze examples.
• Ability to set priorities across teams while staying hands-on. Use of AI tools and agents to multiply output and verify work. Collaborative mindset with researchers, chemists, and robotics engineers.
• Helpful But Not Required
• Experience with chemistry or scientific data, vision-language models, video understanding, or embodied AI.
• Background in teleoperation, demonstrations, simulation, active learning, or human-feedback workflows.
• Experience with RL-based data generation and selection, reward modeling, verifiable tasks, or large-scale synthetic-data pipelines.
• What Success Looks Like
• Warp knows what data to acquire or generate next and can turn it into useful model inputs quickly and reliably.
• New data drives measurable improvements in chemistry reasoning, lab perception, and robot behavior.
• Data collection, generation, and learning loops get more efficient over time, and the shared data foundation becomes easier to use and scale.