Arseni's CV

Experience

June 2021 – Current

Responsible for planning and executing a full machine learning development cycle for an upcoming product. Data collection, data labeling, data cleaning, feature selection, data splitting, metrics and statistics, network architecture selection for EFR32, hyperparameter tuning, evaluation, integration into device, and pre-certification support.
Responsible for projecting and building an automated image data collection pipeline. Performing installations at pilot's properties in Spain for said collection.
Assisting in the development of algorithms for recursive channel-pruning of ResNet/YOLO-architectures in PyTorch for size reduction.
Responsible for the specification of AI hardware protocol requirements for product partner.

Working on Masters thesis "Image classification for improving an IR-signal classification in a self-supervised learning pipeline". Collecting data, labeling data, coding additions to in-house labeling software.

Part of the Cameras & Frontdoor development team. Responsible for modifying hardware by soldering and connecting to relay-cards to facilitate remote automated tests. Responsible for writing automated tests using an in-house Python framework.

August 2019 – June 2021

Assisting in project Advagym by Sony. User testing and bug testing of hard- and software. Distributing and installing the hardware at gyms in Sweden, Denmark, and Germany. Minor software work on string interpretation.

March 2019 – August 2019

Assisting in project Advagym by Sony. User testing and bug testing of hard- and software. Distributing and installing the hardware at gyms in Sweden, Denmark, and Germany.

January 2018 – March 2019

August 2018 – June 2022

Engineering computer science masters programme at Faculty of Engineering (LTH). Specialization - Images and Computer Graphics.
Master's thesis: Image classification for improving an IR-signal classification in a semi-supervised learning pipeline.

June 2016 – June 2018

November 2022 – Ongoing

Implementation and integration of a MobileNetV3-Architecture onto a STM32F7 (ARM Cortex-M7) module to detect humans on low power devices, trained on Visual Wake Worlds dataset. Utilizes TinyML NAS and achieves 6 FPS with optimizations.

May 2021 – September 2021

Peer-to-Peer communicating, shared world coordinate system chess game using two cameras and an Aruco marker. Made in C++ using OpenCV for plane/camera pose estimation and OpenGL for graphics rendering.

February 2020 – August 2020

Eyebrow recognition with overlay suggestion for "perfect eyebrows" for Brazilian company Della&Delle. Created in Python using an open-source library dlib with pre-trained facial recognition model and opencv for display. Uploaded to GitHub and deployed to customers' websites using Flask & Microsoft Azure by third party.