Igor Perić

Building and deploying best in class AI models for real estate market in the USA and worldwide, helping buyers, sellers and intermediaries extract the highest value in the process. Some interesting achievements of our brilliant team:

• We have built a scalable platform for processing billions of images per month in realtime
• We fine-tune hundreds of vision and language models a week
• We deploy models to production APIs daily
• We have curated the highest quality real estate listing database over several years and are using it to provide additional value to our customers in every single API request

I help the team stay efficient and adapt quickly to the ever-changing landscape of daily AI developments. My previous experience in building cutting-edge AI platforms and successfully developing AI healthcare products to run on mobile devices helps me optimize Restb's models to run very efficiently while staying the most accurate real estate models in the world. I normally:

• Lead, manage and provide hands-on support to the AI Engineering team, comprising of software engineers, platform engineers and PhD researchers
• Monitor recent research and industry trends, cut trough the hype, define and implement the AI strategy company-wide
• Identify and implement cost-efficiency measures both on technical and operational level

Driving the development of motion tracking solutions for clinical applications, healthcare and sports industry. Transforming state-of-the-art computer vision and data processing research to fast, interactive and easy to use products and services, available across wide variety of devices and ecosystems.

These are some of the hats I wear:

Software Architecture - As a lead iOS developer of multiple mobile apps, I am making sure the team writes code which is easy to read, use and maintain, no matter how much the design specifications change. I advocate the use of popular development patterns and architectural best practices when analyzing user requirements. In my mind refactoring outdated/poor code is an investment, not a waste of time.

DevOps/MLOps - As a Director of R&D I spearheaded the design and implementation of a toolchain for dataset annotation and automated training of production-ready event detection models. The system I helped build includes cluster management, training job control, accuracy evaluation, model versioning and deployment to production. This toolchain is used not only to speed up prototyping, but to evaluate and compare the models and drive business decisions.

Algorithm engineering - I build efficient algorithms for event detection in a continuous motion data stream, delivered over bluetooth 100 times per second from a proprietary IMU sensor. It involves a lot of trigonometry, linear algebra, some robotics and a lot of engineering. Very frequently I read a lot of papers and adapt open-source implementations of state-of-the-art. I translate the product design specifications into a set of formal, strictly defined tasks and pose expected limitations. The harder the problem is, the bigger the kick I get out of tackling it.

Visit BioMech Leadership page for more details.

As a director of a remote research and development center for an American company I am enforcing efficient and legal conduct of international business between Serbian and USA legal entity. As a part of that I am bridging cultural differences between teams, managing expectations on both ends, continuously identifying talented people to grow the team, conducting interviews, hiring and maintaining a healthy international work environment. I am trying hard to promote an (unfortunately) transformative idea in our region - employee is the most valuable asset a company owns and should be treated with maximum respect, fairness and great care.

In a role of technical lead I am helping a team of developers reach maximum efficiency, growth and satisfaction at workplace by conducting rigorous code-reviews, 1-on-1 walkthroughs, pair programming and dispersion of knowledge across all of the team members. I learned how powerful and beneficial investments in people are in a growing company.

I developed robotics systems for industrial applications and service robotics. I've extensively used Robotic Operating System (ROS) and coded mostly in C++ and Python.

Visit FZI homepage for more details.

I was a part of a Human Brain Project team which built large-scale simulations of virtual robotic environments and agents. We mostly worked on Neurorobotics platform, a system for connecting biologically inspired models of spiking neural networks to physics simulators, real robots and supercomputers capable of simulating large brains in realtime.

The aim of project was to provide neuroscientists a set of powerful and flexible tools to verify the theories about underlying computational principles of the brain.

Visit Human Brain Project - Neurorobotics Platform for more details.

As part of my Master thesis for Computer Vision and Robotics degree I worked on developing a neural network classifier for detecting presence and type of ADHD in patients, based on attention biomarker and eye tracking data.

Visit BrainGaze homepage for more details.

I am member of a team competing in European robotics competition euRathlon 2014 (www.eurathlon.eu) held in September in Spezia, Italy. My role dealing with sensory data processing, autonomous mapping and motion planning algorithms and detection of anomalies using underwater camera image processing (OpenCV). Sparus 2, robot I am working on, is built on Robotics Operating System (www.ros.org) and runs on Linux. I'm developing autonomous system for solving mapping and localisation tasks in both C++ and Python, using third party ROS libraries (octomap for mapping, cola2 AUV control architecture, etc).

I developed ball tracking solutions for high FPS cameras, which include motion detection, ball path prediction, camera calibration and 3D reconstruction. Most of the time I was researching different image processing techniques to find the best approach to perform ball tracking in the most efficient, robust and accurate way.

During 4 years of my membership in Microsoft Student Partner program I have promoted Microsoft development technologies in my country through public speaking, attending and organising various tech events and participating in different types of software development contests.

As a result of two years of research supervised by Prof. Rüdiger Dillmann in modelling cognitive architectures using spiking neural networks for robotic systems, I published six conference papers and one journal paper.

My model of convolutional associative memories using spike-timing mechanics won the best paper award in 2017, being the first paper proposing a way the weight-sharing mechanism from CNNs could be implemented by our biological brains using Fourier transform to store and associate memories.

Visit my ResearchGate profile or portfolio for more details.

Over 4 semesters (120 ECTS), I attended three universities:

1. Université de Bourgogne, France
2. University of Girona, Spain
3. Heriot-Watt University, United Kingdom

I was required to cover the basis of signal and image processing, tools and methods of computer vision (compression, segmentation, real-time, shape recognition, 3D vision), fundamentals of robotics, control and programming for robot autonomy and intelligence, medical imaging, infrared imagery, X-Ray and ultrasound imagery and many others.

Studied fundamentals of information processing theory, basics of computer networking, databases, programming, app development, computer architecture, operating systems and project management.

Finalist of two international student tech innovation competitions Microsoft Imagine Cup 2012 and 2013.

Visit my university home-page and my blog for more details.

Studied basics of electronics, computer science and programming. Won multiple regional and state championships in algorithms and data structures. Participated in finals of two International Olympiads of Informatics (IOI).

Visit my high-school homepage, my IOI profile or my blog for more details.