Research Engineer - Data (AI)

TL;DR

Research Engineer - Data (AI): Build and drive the data foundation for research efforts in materials, energy, and physical sciences with an accent on sourcing scientific datasets, integrating experimental data, and ensuring high-quality inputs for frontier models. Focus on designing scalable pipelines, data quality systems, and tooling for reproducibility and researcher collaboration.

Location: Lab in Menlo Park, prefer located in Menlo Park or San Francisco but flexible based on role.

Compensation: $350,000-400,000 annual base commensurate with experience.

Company

AI and physical sciences company building state-of-the-art models to accelerate breakthroughs across materials, energy, and beyond. Backed by world-class investors and growing rapidly.

What you will do

• Own data strategy across the training stack, identifying gaps and shaping the roadmap with research leads.
• Source, evaluate, and procure external datasets in chemistry, physics, materials science, and more.
• Build pipelines for ingesting, processing, and versioning large-scale heterogeneous datasets.
• Design data quality systems including deduplication, filtering, and normalization at scale.
• Integrate lab experimental data, simulations, and model outputs into the training stack.
• Develop tooling for data inspection, querying, metadata tracking, and reproducibility.
• Collaborate on token budgets, data mixing, and curriculum design with ML engineers.

Requirements

• Bachelor’s degree or equivalent.
• Experience building large-scale data pipelines for LLM pretraining or midtraining.
• Expertise in data quality techniques like MinHash, SimHash, perplexity filtering, and PII scrubbing.
• Work with scientific data formats (papers, patents, simulations, lab exports) and normalization.
• Distributed processing with Spark, Ray, or Dask at TB/PB scale.
• Dataset versioning, lineage tracking (DVC, Delta Lake).
• Strong Python for production tooling; collaborate with ML researchers.
• Research mindset with experiments and iteration.

Nice to have

• Curating scientific datasets for domain-adaptive pretraining or tuning.
• Synthetic data generation and verification.
• Background in physical science or engineering.
• Multimodal data integration (text, numerical, molecular, spectral).

Culture & Benefits

• Visa sponsorship: Yes, with legal support.
• Operate at frontier pace with deep expertise, ownership, and drive.
• Team of top scientists, engineers, and problem-solvers defining the frontier.