Jovan Cicvarić

Machine Learning Engineer

• • Developed and deployed a computer vision model for construction waste classification, achieving high accuracy and improving sorting efficiency.
• • Engineered an object size estimation model using monocular and stereo camera inputs, successfully identifying key objects in a majority of test cases.
• • Designed and implemented the full data labelling pipeline for 80K+ images, and communicated with experts and customers to define annotation requirements.
• • Led the development of a novel plastics classification model by combining multispectral vision with an LLM-based approach, utilizing a newly created dataset from custom multispectral sensors.
• • Conducted comparative studies on classification performance that demonstrated the superior performance of RGB data over multispectral data for this specific task.
• • Developed training tracking and model deployment pipeline using MLFlow and Triton Inference Server.

• • Led full-stack development of the scientific paper recommendation system, scaling the platform to support over 20K new users.
• • Engineered the personalized recommendation system, trained models on user votes against a database of 3 million scientific articles.
• • Optimized SQL queries and the embedding storage system, achieving a 5x improvement in latency and memory usage.
• • Co-authored the creation of a dataset of abstract annotations (2425 sub-sentence labels within 691 abstracts).

Student Assistant

• • Assisted in preparatory university courses in Mathematics and Physics for 10th-grade students.
• • Checked homework, gave personal reviews and personalized study guidance.

M.Sc. Machine Learning

GPA 1.1 (best: 1.0) · Graduated with distinction

Deep Learning · Computer Vision · Reinforcement Learning · Statistical ML

B.Sc. Applied Mathematics & Physics

GPA 4.66 (best: 5.0)

Calculus · Informatics · Linear Algebra · Differential Equations · Computational Mathematics · Physics

Scholar Inbox: Personalized Paper Recommendations for Scientists

ACL Demo, 2025

M. Flicke, G. Angrabeit, M. Iyengar, V. Protsenko, I. Shakun, J. Cicvaric, B. Kargi, H. He, L. Schuler, L. Scholz, K. Agnihotri, Y. Cao, A. Geiger.

Generative Dataset Distillation: A New Hope?

Workshop on the Dataset Distillation Challenge, ECCV 2024

M. Schneider*, J. Cicvaric*, A. Sauer, A. Geiger, K. Chitta.

Generative Dataset Distillation

• • Master Thesis under the supervision of Prof. Andreas Geiger and Kashyap Chitta.
• • Introduced new approaches combining dataset distillation and generative modelling.
• • Worked with ImageNet-1k, CIFAR10/100, StyleGAN2 and StyleGAN-XL.
• • Used generative dataset distillation for imitation learning on CarRacing env — achieved 80+ of original score with just 1 image per class.

Crypto News Telegram Bot

• • Implemented and deployed Telegram Bot for receiving crypto news from Twitter, Discord and CoinMarket.
• • Actively used by 250+ people and providing access to 60+ crypto-related projects.

Laser-hockey RL

• • Implemented DDPG and TD3 agents for laser-hockey environment.
• • Ranked 3rd out of 70+ participants in a tournament by Autonomous Learning Group @ MPI-IS.

CarRacing IL & RL

• • Implemented DQN, Imitation Learning and Modular pipeline with geometric controller for CarRacing environment.
• • Placed 3rd overall with 35+ participants organized by Autonomous Vision Group @ Uni Tübingen.