Member of Technical Staff, Foundation Models (AI for Chemistry)

About the role

We are building foundation models for chemistry and the sciences from scratch. General purpose LLMs do not work well here; the domain requires deep understanding of molecules,

spectra, reactions and lab records, and correctness is decided by physics and experiment, not by text similarity.

This is an individual contributor role; the candidate will work on model building and integrating the same into our product.

Responsibilities:

- Run pretraining and continued pretraining for scientific data. Decide tokenisation, data mixture and curriculum.
- Post-train models for multi step scientific reasoning using RL with verifiable rewards, reward models and long reasoning traces.
- Build synthetic data and distillation pipelines. Teacher sampling, rejection sampling, trace filtering, and distilling large models into smaller ones without losing reasoning quality.
- Train world models for reaction and process dynamics. Learn state representations, predict outcomes, and use these models for search and planning.
- Handle large scale distributed training, including throughput, checkpointing and stability issues.
- Own model evaluation end to end, including benchmarks that do not exist yet for our tasks.
- Publish novel research in top AI and scientific venues.




- Collaborate with synthetic chemists, computational chemists and software engineers.
- Mentor junior researchers and contribute to technical strategy.

Required skills:

- 5+ years of hands-on work in training or post-training large models. Fine-tuning API models alone will not be sufficient.
- Strong command over PyTorch or JAX, and distributed training with FSDP, DeepSpeed or Megatron.
- Depth in at least two areas out of: reasoning and RL post-training, distillation and synthetic data, world models or model based RL, large scale pretraining.
- Good experimental discipline. Should be able to design the right ablation and drop an idea when results do not support it.

Valuable to have:

- Publications or open source work in reasoning, distillation or world models.
- Background in chemistry, computational chemistry, materials or drug discovery.
- Alternatively, a record of picking up a new field quickly.
- Experience with molecular representations such as SMILES, graphs, 3D structures or
- spectra.
- Inference side experience: quantisation, speculative decoding, serving efficiency.

Qualification:

PhD, M.Tech or B.Tech in CS, AI. Minor in Chemistry, Physics or related fields. Equivalent industry experience is also acceptable.

📌 Member of Technical Staff (Bengaluru)
🏢 Mstack AI
📍 Bengaluru