Ml Engineer

Job Overview:

We are looking for a brilliant Member of Technical Staff, Foundation Models (AI for Chemistry) with 4-9 years of core model training experience to build domain-specific scientific foundation models from scratch. In this individual contributor role, you will move past general-purpose LLMs to architect networks that understand molecules, chemical reactions, spectra, and laboratory records where correctness is bounded by physics and experimental verification. This position requires deep, hands-on mastery of large-scale distributed pretraining, advanced post-training reasoning pipelines, and rigorous experimental discipline.

Key Responsibilities:

• Core Pretraining & Architecture: Run large-scale pretraining and continued pretraining setups across massive scientific data blocks; determine optimal tokenization, data mixture profiles, and curriculum learning tracks.
• Advanced Post-Training: Train models for multi-step scientific reasoning patterns leveraging reinforcement learning (RL) with verifiable rewards, custom reward models, and prolonged reasoning traces.
• Data Generation & Distillation: Build automated synthetic data pipelines and sophisticated distillation architectures utilizing teacher sampling, rejection sampling, trace filtering, and high-fidelity model compression.
• World Modeling & Search: Develop and train world models for reaction and chemical process dynamics, learning continuous state representations to accurately predict physical outcomes for complex search and planning loops.
• Distributed Infrastructure & Evaluation: Orchestrate large-scale distributed model training operations to maximize throughput, stability, and checkpointing while engineering custom benchmarks for previously unrated scientific tasks.

Required Skills & Qualifications:

• 4-9 years of direct, hands-on experience training, pretraining, or post-training large-scale deep learning models from the ground up (exclusive reliance on fine-tuning wrapper APIs is insufficient).
• Elite programming command over PyTorch or JAX, accompanied by practical production deployment of distributed frameworks such as FSDP, DeepSpeed, or Megatron.
• Technical depth in at least two structural domains: reasoning/RL post-training, model distillation, synthetic data generation pipelines, world models, or model-based reinforcement learning.
• B.Tech / M.Tech / PhD in Computer Science, Artificial Intelligence, or highly quantitative fields; background or domain minor in Chemistry, Computational Chemistry, Physics, or Materials Science is highly desirable.