Research Scientist, Machine Learning

Onepot•South San Francisco, CA

5d•$200,000 - $250,000•Onsite

About The Position

onepot is automating chemistry with the goal of enabling a self-improvement loop for chemistry by combining AI and advanced robotics. In this loop, AI systems design experiments, robotic systems execute them, and the resulting data improves the next generation of models. This goal can only be achieved by bringing together people from different backgrounds: ML engineers, chemists, computer scientists, and hardware engineers. We are building a small, unusually ambitious team and are looking for a machine learning researcher to join us to help train the next generation of chemistry models. We train models for a wide range of tasks, including reaction planning and outcome, input material costs, and mass spectra prediction. Most of these models are state-of-the-art; many are trained on proprietary datasets that are larger, and higher quality, than what exists in the literature. Our models are primarily deployed internally for real workflows. As such, we maintain a tight feedback loop between usage, data, and model training.

Requirements

Strong fundamentals in machine learning with a deep understanding of modern empirical/experimental ML
Experience developing novel model training techniques or dealing with novel tasks or datasets
Evidence that you can move quickly, make good decisions with incomplete information, and solve difficult problems without waiting for detailed instructions
Experience in a startup, research group, competition team, or other environment where you had significant ownership and limited resources is particularly relevant
Familiarity with PyTorch or other machine learning frameworks
Knowledge of basic machine learning theory
Enthusiasm about working across the entire machine learning stack (data, training, inference, deployment)
Comfort working across disciplines and learning unfamiliar technical areas as necessary
Strong written and verbal communication skills
Curiosity and excitement about chemistry
A strong bias toward building, testing, and learning from real systems
Ability to work extended hours and weekends as necessary
Ability to work safely in an active chemistry laboratory and around scientific equipment. This position does not involve lab work, but some projects may require an understanding of lab workflows.

Nice To Haves

Training LLM models (particularly mid- and post-training)
Computer vision and embedded systems/robotics
Machine learning systems (kernels, distributed training, etc.)
Familiarity with chemistry models (retrosynthesis, mass spec modeling, etc.) or cheminformatics
Active learning or other techniques suited for low-data regimes
Scaling experiments and determining scaling laws
Want to see your models used rather than benchmarked — here the loop closes in the lab, not on a leaderboard
Energized rather than discouraged by novel tasks with no established baseline or dataset
Reach across the whole stack, from data acquisition through training to what runs in production
Move with urgency while keeping enough rigor to know whether a result is real
Want substantial responsibility early, including over what gets built and why
Willing to work outside a narrow job description to make the overall system succeed

Responsibilities

Train the next generation of chemistry models, and enable synthesis of previously inaccessible molecules.
Work with lab staff on data acquisition, train models of many different shapes and sizes, and deploy models directly into experimental workflows.
Build and help maintain infrastructure and abstractions for training a diverse set of models; this includes data pipelines and various system tasks such as parallelism strategies and quantization.
Train models with superhuman chemistry intuition and experimental capabilities.