Design, implement, and optimize RL, imitation learning for a variety of use cases under Drug Discovery and Biotech.
Build scalable, transferable, and production-ready codebases using PyTorch.
Explore and prototype novel learning approaches that push the boundaries of efficiency and adaptability.
Generate intellectual property, publications, and open research collaborations.
Bring cross domain capabilities from machine learning disciplines into generative drug design.
Requirements
M.S. or equivalent experience in Computer Science, Artificial Intelligence, Biotechnology, BioEngineering, or a closely related field
(MTech + 2
years of research experience with a strong publication record
Core Deep Learning & Architecture Design
Deep Proficiency in PyTorch: You possess native fluency in PyTorch. You can write custom training loops, implement complex loss functions, and debug autograd issues without relying on high-level abstractions.
Custom Model Development: Proven track record of designing custom neural network architectures rather than simply fine-tuning pre-trained models. You understand the mathematical intuition behind architectural choices (attention mechanisms, normalization layers, skip connections) and can innovate upon them.
Training Dynamics: extensive experience with the nuances of neural network training, including gradient clipping, learning rate scheduling, mixed-precision training (AMP), and diagnosing convergence issues in deep networks.
Strong theoretical and practical grasp of Model-Free and Model-Based RL.
Experience implementing algorithms such as PPO (Proximal Policy Optimization), TRPO, DPO, DQN, A2C, or SAC (Soft Actor-Critic).
Familiarity with Reinforcement Learning from Human/AI Feedback pipelines to align LLMs or agentic behaviors.
Ability to design custom simulation environments and reward functions that accurately map to complex real-world objectives.
You understand message-passing paradigms, graph isomorphism challenges, and how to scale GNNs to large, heterogeneous graphs.
Deep knowledge of geometric deep learning, specifically designing E(n) or SE(3)-equivariant networks.
You understand how to bake physical symmetries (rotation, translation, reflection) directly into model architectures for 3D data or physical simulations.
Experience scaling training across multiple GPUs/nodes using tools like accelerate, DeepSpeed, FSDP, or Ray.
Proficiency in containerization (Docker, Kubernetes) and model serving frameworks (e.g., vLLM, Triton Inference Server, or TorchServe).
Strong software engineering practices in Python, including testing, CI/CD, and writing modular, maintainable research code.
Self-starter mindset: industrious, independent, and able to generate ideas and drive them forward without waiting for direction.
Excellent problem-solving skills and adaptability to shifting research directions.
Strong collaboration skills, with experience working in interdisciplinary and fast-paced teams.
Tech Stack
Docker
Kubernetes
Python
PyTorch
Ray
Benefits
Ample opportunities to learn, grow and interact with colleagues from varied experience and backgrounds around the globe