Hi, my name is

Cyrus Behroozi.

I'm an AI and computer vision software developer.

I'm a senior software developer based in the United States specializing in high-performance CPU and GPU machine learning inference.

About Me

Hello! I'm Cyrus, a computer vision software developer based in Los Angeles. I studied mechatronics engineering at the University of British Columbia. I previously worked as a senior C++ computer vision engineer at Trueface (a Pangiam company) where I led the SDK team.

Aside from coding, I play the djembe drums and love to travel - I've lived all across South America.

Here are a few technologies I've been working with recently:

C++
Python
x86, ARM
TensorRT, ONNX Runtime
Linux

Where I've Worked

Senior C++ Computer Vision Engineer | SDK Team Lead @ Trueface (A Pangiam Company)

Jun 2019 - Mar 2024

Led SDK and Matcher On-Prem (REST API server in docker container) teams at Trueface, a computer vision company specializing in face recognition.
Spearheaded research effort to achieve number 1 global rank for speed by National Institute of Standards and Technology, Face Recognition Vendor Test.
Wrote high-performance C++ CPU and GPU machine learning inference code which is compiled for X86 and ARM, for Linux, MacOS, Windows, used by DHS, airlines, and airports.
Built thermal-camera embedded elevated body temperature detection solution which was used by US airports during Covid pandemic.
Created product vision and roadmap, attended sales calls, provided client support, wrote core SDK code, wrote and maintained documentation and sample apps, built out CI/CD infrastructure.

C++ Developer Co-op @ DarkVision Technologies

Jan - Aug 2018

Responsible for C++ 17 development for embedded Linux and desktop Windows applications.
Created firmware programming utility to parse Intel HEX file and write packets to flash memory. Reduced the firmware programming time by factor of 5.
Built G-Code programming interface complete with console and text editor to send commands to tool via serial, replacing the existing method of using PuTTY.
Implemented H.265 (HEVC) hardware codec to encode ultrasound data in real-time, significantly decreasing encoding time.

Some Things I've Built

Featured Project

TensorRT C++ API Tutorial

The most comprehensive open-source tutorial on how to use the TensorRT C++ API for high-performance GPU machine learning inference. Includes video recording and presentation slides from live presentation at Venice Computer Vision Meetup.

TensorRT
C++
CMake
OpenCV
CUDA

Featured Project

YoloV8 TensorRT C++

A C++ implementation of YoloV8 object detection and semantic segmentation using TensorRT.

TensorRT
C++
CMake
OpenCV
CUDA

Featured Project

SDK Design Tutorial

Tutorial project and code outlining how to design a language agnostic computer vision SDK for cross platform deployment and maximum extensibility. Includes an associated blog post and video recording of my live presentation at a Venice Computer Vision meetup event.