Ali Vaziri

Experience

From research prototypes to systems at scale.

I build learning, simulation, and control software across the full workflow: data preparation, distributed training, evaluation, inference, and deployment-oriented testing.

Industry research

2024–2026
Summer 2026
Sunnyvale, CA

AI/ML Software Engineering Intern

General Motors · Cruise Team

  • Built an end-to-end multimodal ML system that reconstructs missing vehicle motion in recorded-driving simulations.
  • Created cloud data and featurization pipelines for more than 38,000 highway scenes from approximately 1 TB of raw tracking data.
  • Scaled training across eight H100 GPUs with PyTorch DistributedDataParallel, NCCL, and mixed precision, reducing training time from 12 hours to 3.
  • Reduced best-candidate average position error by 24% relative to the company baseline on about 58,000 held-out trajectories.
  • Reworked streaming evaluation and parallel loading, reducing full-test runtime from more than 60 minutes to approximately 9.
  • Implemented oriented-footprint collision metrics at 10 Hz for ego-to-surrounding and surrounding-to-surrounding vehicle interactions.
Summer 2024
Cambridge, MA

Research Intern · Machine Learning Modeling and Optimization

Mitsubishi Electric Research Laboratories (MERL)

  • Developed a physics-constrained Neural ODE–GRU model for vapor-compression HVAC systems.
  • Improved forecasting accuracy by 14% and runtime by 5.7× relative to the previous modeling approach.
  • Combined continuous-time dynamics, recurrent modeling, and physical constraints to learn from limited data.
  • Delivered reusable PyTorch software for data processing, training, evaluation, and GPU-accelerated control experiments.

Academic research

2023–present
2026–present
Michigan

Graduate Research Assistant

Michigan State University

  • Developing machine-learning methods for high-dimensional dynamical systems and intelligent control.
  • Studying continuous-time optimal control for Neural ODEs and learned models of Burgers and Navier–Stokes systems.
  • Developing GPU-compatible ensemble smoothing methods for spatial and tensor-valued models.
2023–2025
Lawrence, KS

Graduate Research Assistant

University of Kansas

  • Reformulated MPC as Bayesian inference for neural state-space vehicle models; tested configurations solved up to 200× faster than CasADi/IPOPT benchmarks.
  • Designed robust motion planning with Student’s-t distributions, sequential Monte Carlo, and ensemble Kalman smoothing.
  • Trained GRU and ResNet vehicle-dynamics models from real-world driving data and connected them to Bayesian planning and trajectory optimization.
  • Developed matrix-variate ensemble Kalman smoothing with up to 18.6× faster computation and about 11× lower memory use than a vectorized GPU implementation.

Background

Education

2026–presentMichigan

Ph.D., Mechanical Engineering

Michigan State University
Machine learning, stochastic processes, Bayesian inference, and optimal control

2023–2025Kansas

M.S., Mechanical Engineering

University of Kansas · GPA: 4.00/4.00
Heavy-Tailed Bayesian Motion Planning for Autonomous Vehicles

2023–2025Kansas

Ph.D. studies, Mechanical Engineering

University of Kansas
Transferred to Michigan State University

2016–2021Tehran

B.S., Marine Engineering

Sharif University of Technology · GPA: 3.55/4.00
Collision-Free Marine Waste-Collection Robot

Technical toolkit

Methods and systems

Machine learning & modeling

Deep learning, generative and diffusion models, multimodal and sequence modeling, CNNs, recurrent networks, Neural ODEs, probabilistic modeling, physics-guided learning, and system identification.

Control, planning & estimation

Stochastic processes, model predictive control, Bayesian inference, inverse problems, trajectory optimization, robust motion planning, Kalman filtering and smoothing, particle filtering, and sequential Monte Carlo.

Programming & ML systems

Python, C/C++, CUDA, PyTorch, TensorFlow, NumPy, SciPy, DDP, NCCL, mixed precision, distributed GPU computing, profiling, and testing.

Simulation, cloud & robotics

Autonomous-vehicle simulation and evaluation, Google Cloud Storage, Dataflow, memory-mapped datasets, ROS, MuJoCo, PyBullet, CasADi, IPOPT, MATLAB/Simulink, Bazel, Linux, and Git.