Deepak R.Sastry

Machine learning engineer with an M.Sc. in reinforcement learning from the University of Alberta.

Experience building end-to-end training systems in PyTorch: custom simulators, RL training pipelines, distributed jobs on SLURM clusters, and experiment tracking, applied to industrial scheduling with PCL Construction and to FPGA placement. Publications at AAMAS 2025, ISEDA 2024, and the NeurIPS ML for Systems workshop.

Seeking Machine Learning

Engineer and Research Engineer roles.

Research Engineer, Reinforcement Learning for Construction Scheduling (M.Sc. Thesis) at IRL Lab, University of Alberta with PCL Construction (2025-04 – Present)

• Built a custom simulator for a live PCL construction project and the end-to-end training pipeline around it, encoding crew movement cost, labour-curve smoothness, and total project duration as the agent objective on a 3D project grid.
• Trained MaskablePPO agents with action masking to enforce precedence and resource constraints, running distributed training jobs on Compute Canada clusters through SLURM with experiment tracking in Weights & Biases.
• Measured agent schedules against professional human planners in a controlled study on a project subarea: labour curves ~60% smoother than the planners, with schedule duration within ~6–10% of the strongest baseline and shorter than the planners.

Deep Reinforcement Learning for FPGA Placement at IRL Lab, University of Alberta (2023-06 – 2024-01)

• Built the first deep RL agent for FPGA placement: PPO with a ResNet and Graph Attention Network state encoder plus invalid-action masking, matching the average wirelength of the state-of-the-art VTR placer on 15-block placements, outperforming it in individual runs.
• Added Random Network Distillation for exploration, significantly improving final wirelength (p < 0.03), and restructured training into a divide-and-conquer scheme that cut average wirelength by up to ~6% against the monolithic RL baseline.

AI Career Accelerator Program at Alberta Machine Intelligence Institute (Amii) (2024-01 – 2025-12)

• Scoped ML feasibility for two client engagements, an industrial hydraulics and pipeline application and an interviewer to interviewee matching system, working with Amii scientists to deliver technical recommendations to the clients.

M.Sc. in Computing Science – University of Alberta (2024-01 – 2026-06)

B.Sc. in Computing Science (Specialization) – University of Alberta (2019-09 – 2023-12)