Rouzbeh Shirvani

Mountain View, CA

Hi, I’m Rouzbeh, I am an ML engineer and researcher focused on building advanced ML systems—from model design and training to production software and efficient deployment on real hardware. My work covers transformers, vision, speech, model optimization, and deployment tooling. I’m most drawn to problems where building a strong model is only one part of the job, and the real challenge is making it fast, reliable, and useful in practice.

At Syntiant, I’ve worked across model training, optimization, embedded inference, and deployment tooling. I developed the company’s first transformer-based search system, now running on millions of cameras globally, and designed and built the NDP Model Converter, the core software library used to deploy ML models across Syntiant’s neural processor family. Earlier, at Athelas, I led the full ML stack for a platelet-monitoring device across model development, hardware, software, and regulatory testing.

What I do best is turn technically ambitious ML ideas into systems that actually ship. That means training the models, building the surrounding software, optimizing for real constraints, and carrying the work far enough that it becomes useful in the world.

Rouzbeh Shirvani

Work Experience

Staff ML Engineer / Tech Lead

Syntiant — Mountain View, CA  (+Two Acquisitions)

2021 – Present

  • Training and implementation of Syntiant's first Transformer based search model. A model that is able to perform search across a fleet of devices. Today it is deployed on millions of cameras globally.
  • Led the design and implementation of Syntiant's NDP Model Converter software package. Today it is the core software library that enables millions of devices globally to run ML models on Syntiant's NDP chip family.
  • Designed Syntiant's first Transformer-based embedded search product for multi-camera embedded systems.
  • Designed and implemented the first vision-based model for the NDP200 chip; trained and ported the first ASR model for Syntiant's NDP family.
  • Contributed technical diligence during two acquisitions and partnered with leadership and external stakeholders on next-generation chip specifications.

ML Engineer

Athelas (merged with Commure) — San Jose, CA

2019 – 2020

  • Led the end-to-end ML program for platelet-monitoring devices, spanning model development, PCB design, dashboard software, and FDA-oriented testing.
  • Built and deployed production ML models for blood-cell / platelet detection.
  • Designed the ML stack and first PCB for the platelet-monitoring device.

Compiler Engineer (GPU)

Qualcomm — San Diego, CA

2018 – 2019

  • Compiler optimization for Graphics Processing Units.

ML Engineering Intern

Adobe Sensei — San Jose, CA

Summer 2018

  • Built an ML-based recommendation system for Photoshop users using behavioral interaction history
  • Received a return full-time offer

Research Intern, Siri

Apple — Cupertino, CA

Summer 2017

  • Built a reinforcement-learning-based system to improve user interactions with Siri

Research Engineer (Funded Project)

Intel — Autonomous Drones

2017

  • Led an Intel-funded autonomous drone project focused on object avoidance using radar and vision

Research Intern

Massachusetts Institute of Technology (MIT)

Summer 2016

  • Analyzing the behavior of mice under certain conditions. The task included implanting electrodes into mice brains and studying their behavior by analyzing neural signals and responses

Founder & CEO

Value Cumulation

2021 – 2023

  • Founded and built a financial-insights SaaS product solo, then grew the team to include design, engineering, and sales. Operated for nearly two years before winding down due to market saturation.

Education

PhD, Computer Science and Electrical Engineering

Howard University — Washington D.C., USA

2014 – 2018

Master of Science, Electrical Engineering (Electronics)

Sharif University of Technology — Tehran, Iran

2010 – 2012

Bachelor of Science, Electrical Engineering (Electronics)

Iran University of Science and Technology — Tehran, Iran

2006 – 2010

Selected Papers & Posts

  • NEWImage Generation Models: A Technical History [Link]
  • Large Language Models: A Decade of Progress [Link]
  • Optimizing Neural Networks [Link]
  • Machine Learning Modeling for the Edge Computing Era [Link]
  • Pixels to Play: Mastering Atari with Deep Learning [Link]
  • Q-Learning and Bellman Equation [Link]
  • Rouzbeh Shirvani. “Image Generation Models: A Technical History.” arXiv preprint arXiv:2603.07455 (2026). [arXiv]
  • Gloria Washington and Rouzbeh A. Shirvani. “Towards Understanding and Modeling Empathy for Use in Motivational Design Thinking.” arXiv preprint arXiv:1907.12001 (2019). [arXiv]
  • Siamak Aram, Rouzbeh A. Shirvani, Eros G. Pasero, and Mohamed Chouikha. “Implantable Medical Devices; Networking Security Survey.” Journal of Internet Services and Information Security (JISIS), Vol. 6, No. 3, pp. 40–60, Aug. 2016. [PDF]
  • Rouzbeh A. Shirvani, Mario Piergallini, Gauri S. Gautam, and Mohamed Chouikha. “Word-Level Language Identification and Predicting Codeswitching Points in Swahili-English Language Data.” Conference on Empirical Methods in Natural Language Processing (EMNLP), Austin, Texas, USA, Nov. 2016. [arXiv]

For a full list of posts refer to the Posts page. For a full list of papers refer to my Google Scholar page.