Chai Discovery
Research Engineer
About this role
Chai Discovery seeks a Research Engineer to develop and optimize machine learning frameworks for molecular design. You'll build core ML infrastructure, evaluation systems, and distributed training stacks that power drug discovery at scale for leading pharmaceutical companies.
What you'll do
- Develop agent-powered auto research frameworks to improve model training and inference speed
- Build evaluation frameworks, datasets, and algorithms to systematically assess model performance
- Analyze ML model failure modes and collaborate with Research Scientists on mitigation experiments
- Own the distributed ML training stack and eliminate bottlenecks at architecture and kernel levels
- Optimize GPU clusters and large-scale model training infrastructure
- Improve ML workload performance through parallelism, quantization, and kernel optimization
What they're looking for
- Python
- PyTorch or JAX
- Software system design
- Distributed ML training
- GPU/CUDA optimization
- ML evaluation and benchmarking
- Kernel optimization (Triton)
- Model performance analysis
Benefits
- Work on frontier AI and biology research with real-world drug discovery impact
- Join a world-class team with high-velocity, ownership-driven culture
- Competitive compensation aligned with market rates
- Opportunity to influence ML infrastructure at scale
Opens the application — the Jobs AI extension fills it for you. Set up autofill
Opens the official application on the employer’s site. No login required.
Chai Discovery
Chai Discovery builds AI-powered tools for molecular biology and drug discovery, enabling scientists to design new therapeutic molecules through advanced AI systems. The company is hiring Security Engineers, Design Engineers, AI Research Engineers, Infrastructure Engineers, and Product Software Engineers to develop and scale its platform.
View all jobs at Chai DiscoveryLikely interview questions
- Describe a time you optimized an ML training pipeline—what bottleneck did you identify and how did you resolve it?
- How have you approached designing evaluation frameworks to systematically measure model performance?