Kristjan Kongas
Machine Learning Engineer at Veeve Inc.- Claim this Profile
Click to upgrade to our gold package
for the full feature experience.
Topline Score
Bio
0
/5.0 / Based on 0 ratingsFilter reviews by:
Credentials
-
Improving Deep Neural Networks: Hyperparameter Tuning, Regularization and Optimization
CourseraFeb, 2018- Sep, 2024 -
Convolutional Neural Networks
CourseraJan, 2018- Sep, 2024 -
Neural Networks and Deep Learning
CourseraJan, 2018- Sep, 2024 -
Structuring Machine Learning Projects
CourseraDec, 2017- Sep, 2024
Experience
-
Veeve Inc.
-
United States
-
Software Development
-
1 - 100 Employee
-
Machine Learning Engineer
-
Jan 2023 - Present
-
-
-
Justacart
-
Estonia
-
Computer Hardware Manufacturing
-
1 - 100 Employee
-
Co-Founder
-
Feb 2022 - Jan 2023
Built a prototype of a smart shopping cart, with automatic product recognition from cameras: * Built a large portion of the algorithmic and device handling parts of the software, such as: product recognition, 3d localization of the product using multiple cameras, handling cameras, weight sensors, and a dedicated QR code scanner from Linux/Python. * Designed and 3d printed plastic parts of the cart, and built the electronics Justacart scored Top 9 in Ajujaht 2022 and made it to Y Combinator W23 batch interview. Show less
-
-
-
Timbeter
-
Estonia
-
IT Services and IT Consulting
-
1 - 100 Employee
-
Deep Learning Engineer
-
Mar 2017 - Apr 2022
Developed, maintained and deployed deep learning based computer vision projects for timber pile images. Major successful projects: * Detecting log ends from images of timber piles. Decreased detection error rate 3x and detection latency by 6x. The resulting multi-stage algorithm consists of several different neural networks and custom filters. * Identical pile search algorithm: search for pictures made of the same pile. This is quite hard, as 1) pictures often look very similar, but are in fact of a different pile, 2) lighting/view angle/weather (eg. snow) can differ a lot between pictures, 3) an overlap might be partial (eg. some of the pile has been removed), 4) false positive rate must be very low to make the system usable * Detecting truck loads from pictures made from side views of trucks * Improved stitching of several parallel images of the same pile into one image Show less
-
-
Education
-
Tallinn Secondary Science School