Founding AI Engineer

We use AI to mine discarded experimental data and drive scientific breakthroughs. Working alongside labs, we discover the materials that will power the new Industrial Revolution.

90% of experiments never make it to publication. Failed runs, abandoned hypotheses, and routine characterization data live in scientists' heads and scattered notebooks. 83 Sciences captures that hidden data at the source, structures it into a queryable "lab brain," and puts it to work: helping researchers learn from their lab's full history and discover new materials.

We are working hands-on with our first cohort of university lab partners and we're backed by Y Combinator (S26). Founded by scientists who lived the file drawer problem firsthand — we're a small team where everyone ships product and talks to researchers directly.

We use AI to mine discarded experimental data and drive scientific breakthroughs. Working alongside labs, we discover the materials that will power the new Industrial Revolution.

About 83 Sciences

83 Sciences (YC S26) is the intelligence engine powering the future of research and materials discovery. Most experimental data (failed runs, unpublished results, raw instrument output) never gets captured. We turn raw lab signals into novel discoveries: capturing and structuring experimental data, shortening research processes, and surfacing the insights that drive new materials.

The role

We're hiring a founding AI engineer to help design insight and discovery extraction models and ship tools for scientists. You’ll own the architecture and build the ML stack alongside our Chief Science Officer and work directly with the founders.

What you'll do

• Ship enterprise-ready products across our platform: data capture, structured experimental records, querying, and analysis tools
• Build our AI systems: multimodal pipelines (vision models for handwritten notebook pages and drawn structures, speech-to-text at the bench), agents that reason over a lab's full experimental history, and models that predict outcomes and propose optimized process conditions
• Work directly with research partners; watch scientists use what you built, then improve it
• Help shape model architecture, data pipelines, and technical direction as an early team member
• Build the data engine that turns messy experimental data (synthesis notes, PXRD, characterization) into training-grade datasets

What we're looking for

• 1+ years of experience managing technical projects at startups and team members or big companies / research labs (e.g., FAIR Chemistry, Google DeepMind, Microsoft Research, OpenAI, Anthropic Lila Sciences, MIT/Stanford/Berkeley/CMU/UToronto AI-for-science groups, or similar)
• Deep in at least two of: geometric deep learning (E(3)/SE(3)-equivariant GNNs), generative models (diffusion, flow matching), ML interatomic potentials
• A track record of shipping enterprise-ready products to real users, end to end, with FDE / customer facing technical experience a plus
• Comfort across the stack: you can get a feature all the way out the door & bias toward speed and ownership in a small, fast-moving team

Logistics: NYC in-person (negotiable for the right person). US work authorization required.

Nice to have: model architecture/fine-tuning experience; background in chemistry, materials science, or scientific tooling; familiarity with scientific data (spectra, diffraction patterns, instrument output), A first-author or major-contributor paper in ML-for-science, geometric ML, molecular/crystal generation, scientific agents, interatomic potentials, synthesis planning, or active learning.

Our platform, Dalton, turns messy, heterogeneous lab data into structured, queryable knowledge. The technical problems we're working on include:

Multimodal data capture: OCR pipelines that convert handwritten lab notebook pages into structured entries and speech-to-text tools

Scientific data analysis: automated PXRD/characterization analysis: phase prediction, peak identification, anomaly detection

Chemistry-specific ML: models tuned on a lab's full experimental history (including failed runs) to predict synthesizability and reaction outcomes and suggest optimal process conditions

You'll work across the full stack (data pipelines, ML, and product) with real lab data and direct feedback from working scientists.