Postdoctoral Researcher · UBC

Dr. Hadi
Keramati

// generative_ai_for_engineering_design.py

I build AI systems that design — fusing diffusion models, reinforcement learning, and physics-based simulation to automate and reimagine how complex engineering systems are created. From ship hulls to drug molecules, I explore the frontier where generative AI meets the physical world.

453
Citations
8
h-index
22+
Publications
1
US Patent
reward_diffusion.py
# Guided diffusion for inverse design
class RewardDiffusion(nn.Module):
  def __init__(self, design_space):
    self.score_net = UNet(design_space)
    self.reward_fn = PhysicsSimulator()
  
  def sample(self, constraints):
    # Denoise toward high-reward region
    x = torch.randn(self.shape)
    for t in self.timesteps:
      ∇r = self.reward_fn.grad(x, t)
      x = self.step(x, t, ∇r)
    return x # optimized design
 
# Applications: heat sinks, ship hulls,
# drug molecules, aerospace structures
 
model = RewardDiffusion("3D").train()
Research Vision

Engineering meets
Generative Intelligence

Inverse Design via Diffusion

Reframing engineering design as a guided generative process — using reward-directed diffusion models to navigate complex, constrained design spaces toward physically optimal solutions.

Physics-Aware AI

Embedding thermodynamic constraints, fluid dynamics, and structural mechanics directly into neural architectures — surrogates and samplers that respect the laws governing the physical world.

Reinforcement Learning for Design

Training autonomous agents that iteratively refine engineering geometries, from heat exchangers to offshore jackets, using multi-agent cooperative RL and shaped reward landscapes.

Cross-Domain Transfer

Developing foundation models for design that transfer across domains — marine engineering, semiconductor cooling, molecular discovery — by learning universal geometry-performance relationships.

Career Journey

From fluid mechanics
to generative machines

2025 — Present
Postdoctoral Researcher
University of British Columbia · Vancouver, BC
Building end-to-end generative 3D design pipelines by fine-tuning and guiding foundation diffusion models. Applications span marine engineering, semiconductor design, aerospace structures, and molecular drug discovery. Introducing GPU-efficient inference-time optimization via geometric particle expansion.
Diffusion Models Ship Hull Optimization Drug Discovery Foundation Models
2022 — Present
VP of Research
Magnative AI
Co-founded and led research at an AI startup commercializing physics-based generative design. Secured $250K in funding, deployed parallelized deep RL pipelines on AWS, and prototyped an LLM-based interaction layer for engineering design software.
Deep RL AWS / SageMaker LLM Integration $250K Funded
2022 — 2024
Postdoc / Entrepreneur in Residence
University of Waterloo
Commercialized a physics-based AI software package with customers worldwide. Secured NSERC I2I grants. Led development of a 3D mathematical design pipeline and supervised a cross-functional team including postdocs, PhD students, master's students, and interns.
NSERC I2I Commercialization Team Leadership
2018 — 2022
PhD Research — 4.0 GPA
University of Waterloo
Pioneered generative thermal design using boundary representation and deep reinforcement learning. Awarded the President's Graduate Scholarship (2 consecutive years). Thesis work became the seed for a US patent and multiple high-impact journal papers.
Deep RL Thermal Design US Patent PGS Award
2017 — 2018
Research Assistant
University of Maryland, College Park
Designed and experimentally tested additively manufactured heat exchangers for aerospace applications (ARPA-E $1.8M project with Boeing). Operated DMLS metal 3D printing systems and won best TA award.
Metal AM ARPA-E Boeing Aerospace
Selected Publications

22 papers, 1 patent,
453 citations

11
Inference-Time Reward Alignment of Diffusion Models via Geometric Particle Expansion
Keramati, Taghvaei, Jaiman · Transactions on Machine Learning Research · 2026
Under Review
1
A Reward-Directed Diffusion Framework for Generative Design
Keramati, Kirchen, Hannan, Jaiman · Engineering Applications of Artificial Intelligence · 2026
Journal
2
HeatGen: A Guided Diffusion Framework for Multiphysics Heat Sink Design Optimization
Keramati, Sadeghi, Jaiman · International Journal of Heat and Mass Transfer · 2026
Journal
12
A Parametric Diffusion Model for Localized Generative Optimization of Ship Hulls
Keramati et al. · Ocean Engineering · 2026
Under Review
5
Deep Convolutional Surrogates and Freedom in Thermal Design
Keramati, Hamdullahpur · Energy and AI · 2023
Journal
6
Deep Reinforcement Learning for Heat Exchanger Shape Optimization
Keramati, Hamdullahpur, Barzegari · Int. J. Heat and Mass Transfer · 2022
Journal
13
System and Method for Heat Exchanger Shape Optimization
Keramati, Hamdullahpur · U.S. Patent No. 11,995,380 · Issued May 28, 2024
US Patent
15
Accelerated Portfolio Optimization and Option Pricing with Reinforcement Learning
Keramati, Jazayeri · ICLR 2025 Workshop on Advances in Financial AI
Conference
16
Reward-Guided Diffusion Model for Data-Driven Black-Box Design Optimization
Keramati, Jaiman · ICLR 2025 Workshop on Deep Generative Models
Conference
17
A Novel Ship Design Optimization Framework Using Fine-Tuned and Reward-Directed Diffusion Model
Keramati, Kirchen, Jaiman · ASME OMAE 2025
Conference
19
Generative Thermal Design Through Boundary Representation and Multi-Agent Cooperative Environment
Keramati, Hamdullahpur · ICML 2022 Workshop on ML in Computational Design
Conference
7
Analytical Solutions for Thermo-Fluidic Transport in Electroosmotic Flow Through Rough Microtubes
Keramati et al. · Int. J. Heat and Mass Transfer · 2016
Journal
Technical Expertise

Tools of the
trade

Machine Learning

PyTorch TensorFlow JAX Diffusion Models Transformers Scikit-learn XGBoost

Reinforcement Learning

Ray RLlib Stable-Baselines3 OpenAI Gym Multi-Agent RL PyTorch RL PPO / SAC / DDPG

Languages & Platforms

Python C++ SQL AWS (EC2/S3/SageMaker) GCP Docker FastAPI

Physics Simulation

OpenFOAM FEniCS Firedrake FreeFEM LBM OpenCASCADE

MLOps & Infrastructure

W&B MLflow Airflow Ray Optuna Linux / Unix

Manufacturing & Robotics

Metal AM / DMLS ROS Arduino CAD G-code 3D X-ray Imaging
Honors & Recognition

A record of
achievement

President's Graduate Scholarship (PGS)
University of Waterloo · 2020–2022
ICLR 2025 Travel Grant Award
RBC Borealis AI · 2025
ASHRAE Grant-in-Aid Award
ASHRAE · 2018–2019
Best Teaching Assistant Award
Dept. of Mechanical Engineering, UMD · 2017
Best Thesis Nomination — Kharazmi International Festival
Iran · 2014
First Rank, Graduation Class — Departmental GPA Record
Babol Noshirvani University · 2012
Research Project Award
Iran's National Elite Foundation
ARPA-E IDEAS Program ($1.79M, Boeing + UMD)
Contributor · 2017–2018

Let's build the
future of design

I'm open to collaborations on generative AI for engineering, industrial partnerships, and conversations about the frontier of physics-informed machine learning.