Looking for a job

Milan Radovanovich

@milanradovanovich

Message

I build production ML, retrieval, and real-time AI systems at scale.

Slovakia

Message

I've built production ML systems for StealthEdge Analytics, WestsideBehavioral, and Antimo, spanning real-time risk and pricing, personalized recommendations, and B2B search. My work has improved predictive precision by 23%, recommendation outcomes by 23%, and search NDCG@10 by 19%.

I own the full ML path—from Kafka, Spark, and feature pipelines to low-latency inference, evaluation, monitoring, and automated fallback. I also build LLM, RAG, and retrieval systems over large document collections, with a focus on measured quality, latency, cost, and reliable production operation.

Experience

Work history, roles, and key accomplishments

Senior ML Engineer

StealthEdge Analytics

May 2025 - Mar 2026 (10 months)

Architected end-to-end real-time ML system processing 500K+ events/second, owning full lifecycle from feature pipelines through model inference and automated retraining. Designed ranking and scoring models (XGBoost + LSTM ensemble) over high-dimensional time-series feature spaces; improved predictive precision by 23% and downstream decision quality by a Sharpe ratio gain of 0.41.

Python XGBoost LSTM Kafka Airflow Spark MLFlow A B Testing Model Monitoring Drift Detection AWS

Machine Learning Engineer

WestsideBehavioral

Dec 2024 - May 2025 (5 months)

Built and deployed a production recommendation system (collaborative filtering + content-based hybrid) for 50K+ users, improving engagement and outcome scores by 23% vs. rule-based baseline. Developed user understanding models: modelled user preference embeddings from behavioral signals (session history, dwell time, interaction patterns) to power personalized content ranking.

Python Collaborative Filtering Content Based Filtering AWS Sagemaker Spark fastAPI Docker AWS ECS Optuna NDCG@K A B Testing

Machine Learning Engineer

Antimo

Jun 2022 - Nov 2024 (2 years 5 months)

Built NLP-based ranking models (fine-tuned BERT on 2M query-document pairs) for search relevance; improved NDCG@10 by 19% over TF-IDF baseline. Developed content understanding representations: entity extraction, topic classification (100+ categories), and semantic embeddings over a 100M+ document corpus using distributed Spark + HuggingFace pipelines.