AI Researcher – Founding AI Team (Chemistry)

CGS; San Francisco (CA); On-site; USD 180,000 - 260,000; Full time

Job summary

CGS in San Francisco is hiring an AI Researcher to help build the founding AI team focused on chemistry applications. You will lead large-scale model training, create training and post-training pipelines, and develop reinforcement learning systems that leverage real laboratory outcomes.

This hands-on role requires collaborating with chemists and product engineers to translate AI research into deployable models for synthesis planning.

Qualifications

• 5+ years of relevant experience, including doctoral research.
• Hands-on distributed training experience using frameworks like FSDP, DeepSpeed or Megatron.
• Strong Python skills and experience with PyTorch or JAX.

Responsibilities

• Lead large-scale model training and post-training experiments.
• Develop training pipelines including pretraining, fine-tuning and distillation.
• Build reinforcement learning systems that learn from experimental feedback.
• Collaborate with chemists and product engineers to connect AI research with chemistry workflows.

Skills

Distributed training

Python programming

PyTorch or JAX

Reinforcement learning

Experimentation discipline

Education

PhD or doctoral research

Tools

FSDP

DeepSpeed

Megatron

Job description

AI Researcher – Founding AI Team (Chemistry)

Working model: Full-time, in-office (5 days a week)

Working hours: 9 AM – 5 PM

Experience: 5+ years (doctoral research experience counts)

Visa sponsorship: Available for US citizens, Green Card holders, H-1B, TN and O-1 (including transfers and renewals)

About the company

A Series B, AI-native chemical manufacturer based in San Francisco is using AI, advanced laboratories and a global manufacturing network to serve more than 100 enterprise customers.

The company is expanding its founding AI team and is looking for an AI Researcher who wants to work at the intersection of frontier machine learning and chemistry.

The role

You will build the training, post-training and reinforcement learning systems that turn chemistry data and real-world laboratory outcomes into reliable models for synthesis planning.

This is a hands‑on research and engineering role, taking ideas from experimentation through to deployed models. You will work closely with chemists and product engineers to solve challenging problems using real experimental feedback.

A chemistry degree is not required. Demonstrated technical contributions and relevant research experience matter more than a particular academic or career path.

What you'll do

• Lead large-scale model training and post-training experiments.
• Develop continued pretraining, fine‑tuning, synthetic data and distillation pipelines.
• Build reinforcement learning systems that learn from experimental feedback.
• Design benchmarks, evaluations and reproducible experiments.
• Run distributed training using frameworks such as FSDP, DeepSpeed or Megatron.
• Investigate model performance, identify failure modes and improve reliability.
• Translate ambiguous research problems into dependable models used in real‑world applications.
• Collaborate with chemists and product engineers to connect AI research with practical chemistry workflows.

What we're looking for

• 5+ years of relevant experience, including doctoral research.
• Personal ownership of large-model training or post‑training experiments. Supporting someone else’s runs is not sufficient.
• Hands‑on distributed training experience using FSDP, DeepSpeed or Megatron.
• Strong Python skills and experience with PyTorch or JAX.
• Deep expertise in at least one of the following:
• Reinforcement learning and reasoning.
• Synthetic data generation and distillation.
• World models and planning.
• Strong experimental discipline, including reproducible evaluations, ablations and reliable baselines.
• Ability to take research from initial exploration through to practical deployment.
• Comfortable working full‑time in person in San Francisco.

What makes you stand out

• Experience with RL using verifiable rewards, sparse feedback or delayed feedback.
• Familiarity with molecular or laboratory data.
• Experience with uncertainty modelling or active learning.
• Evidence of identifying data leakage, reward exploitation or distribution shift.
• Publications, open‑source contributions or experience building benchmarks and synthetic‑data pipelines.
• A track record of turning complex research into models that people actually use.
• Join one of the first members of a founding AI team within an established Series B company.
• Work on frontier AI research with direct applications in chemistry and synthesis planning.
• Own meaningful parts of the training and post‑training strategy.
• Collaborate directly with researchers, chemists and product engineers.
• See your work move beyond research papers into real‑world models and industrial applications.
• Competitive compensation and meaningful equity.
• Founder interview (60 minutes): Motivation, research background, reasoning and interest in building AI for science.
• Technical interview with the Chief AI Officer (60 minutes): Previous training and post‑training work, foundation models for chemistry, evaluation and learning from experimental feedback.
• Research and collaboration interview (60 minutes): Research judgement, execution and ability to take ambiguous problems through to reliable deployment.
• Reference screening (30 minutes): References carry significant weight in the decision‑making process.

At least one interview conversation will take place in person.