Models that ship. Teams that deliver.

Architecture & leadership

• Led the design of end-to-end AI systems: architecture blueprints, technical specifications and integration plans that ensured scalability, security and performance in production.
• Took 3+ AI projects from concept to deployment, defining scope, objectives and deliverables with product teams and client stakeholders.
• Guided the development team technically, and evaluated and selected AI tools and frameworks for the company's stack.

Engineering

• Built vision-guided robotic pick-and-place for objects of varying shapes and structures.
• Developed camera-robot calibration methods.
• Built visual inspection and anomaly detection systems for manufacturing quality control.
• Developed LLM-based text and voice interfaces for natural-language human-robot interaction.
• Built a synthetic data pipeline that combines realistic 3D rendering with generative-AI post-processing, reducing the need for real labeled data.

Engineering

• Led the architecture and hands-on development of transformer-based 3D point cloud segmentation models for autonomous driving and mapping.
• Optimized deep learning models for hardware acceleration using quantization, pruning and operator fusion, achieving 10× faster inference with less than 3% accuracy loss on FPGA targets.
• Contributed to the ML compiler that maps neural networks onto custom FPGA architectures, enabling deployment of models on the hardware.
• Designed training and evaluation data pipelines for large-scale 3D point cloud datasets.

Leadership

• Built and led a team of 5 AI scientists and engineers, covering hiring, mentoring, goal-setting, and code and research reviews.
• Owned the AI technical roadmap: selected, scoped and prioritized AI projects to balance research ambition against product and hardware constraints.
• Partnered with product and hardware teams to turn research results into product capabilities and customer demos.
• Shaped the company's AI research agenda.

Leadership & strategy

• Defined data and ML strategy across 5 business domains, turning business challenges into a prioritized portfolio of data products.
• Built and ran the Digital Data Lab as the company's in-house center for ML delivery, leading a team of 10+ data scientists and engineers through 20+ projects from concept to production.
• Worked directly with business stakeholders to translate requirements into ML solutions and find new opportunities, growing the portfolio to more than 20 projects.
• Directed partnerships with external vendors and academic institutions to speed up R&D.

Delivered ML portfolio

• Retail: customer churn prediction, segmentation and product recommendation.
• Geology: lithology facies classification and grain size distribution prediction from well data.
• Production & refining: predictive maintenance for electrical submersible pumps (ESPs) and multi-stage centrifugal pumps.
• Seismic & remote sensing: GAN-based seismic data inpainting (EdgeConnect), plus computer vision for understanding aerial drone terrain imagery.
• Designed and set up the lab's complete on-premise compute infrastructure.

Research leadership

• Led the research teams at the Belgrade R&D center (15+ scientists and engineers), covering hiring, mentoring and project reviews.
• Defined the center's research strategy and roadmap, choosing research directions that balanced technical ambition against cost and delivery timelines.
• Turned research results into production models for the core product, working with engineering teams in Ireland and Romania.

Computer vision research & engineering

• Developed one-shot and few-shot learning methods (CNN-based) for object recognition and detection, letting the system recognize new retail products from fewer than 20 images per product instead of large labeled datasets.
• Applied weakly supervised learning to object detection, cutting manual annotation effort by 20% by moving from bounding boxes to image-level labels.
• Built synthetic data generation pipelines to augment training data for rare products and edge cases.

Leadership

• Built and led the deep learning computer vision team (3 engineers), owning the roadmap from problem selection to production deployment.
• Introduced deep learning to the company's data science practice, moving image analysis from manual or rule-based processes to production CNN models.

Computer vision

• Built a car make / model / generation recognition system covering 1k vehicle classes, used to auto-tag and validate 100M listings.
• Developed models for car orientation detection and transmission type recognition from listing photos, enabling automated listing quality checks and enrichment.
• Built salvage car detection to flag damaged vehicles in listings, improving marketplace trust and data quality.
• Implemented VIN recognition (OCR) from photos, automating vehicle identification.
• Designed a visual search system that lets users find cars by image similarity.

Pricing & anomaly detection

• Built car price prediction models and anomaly detection for flagging mispriced or fraudulent listings.

• Developed mutual-information-based feature and instance selection algorithms for time-series forecasting, applied to electrical load forecasting, stock market trend prediction and heart disease prediction.
• Published in peer-reviewed journals including Neurocomputing, Entropy and Energies (200+ citations across my research), and contributed to 2 national research projects.