Led AI development of a fraud prevention system deployed in IKEA and Netto, increasing profitability by 0.5%:
Trained real-time Object Detection and Feature Extraction models enabling accurate recognition of 15,000+ products
Developed a two-stage tracking model combining Gradient Boosting and Graph Neural Networks (GNN), using geometric, appearance, and interaction features
Built tools to assess annotation quality and maintain dataset consistency, accelerating data preparation and improving model accuracy
Automated pipelines for video ingestion, annotation tasks creation, model fine-tuning, and deployment, facilitating rapid production rollouts in new stores
Designed and trained a combined multi-task AI model integrating attribute detection, classification, and segmentation capabilities enabling the detection of over 120 chest X-ray pathologies
Co-authored research published in The Lancet Digital Health demonstrating substantial improvement in radiologists' diagnostic accuracy across diverse clinical findings
Integrated into 250+ Australian radiology clinics, the system has improved diagnostic accuracy by 45% and reduced radiologist workload by 20% of cases through automated reporting
Implemented an advanced Integrated Planning System, setting data-driven sales targets and compensation plans for 3,000+ branches, resulting in an annual revenue increase of 8% (KZT 1.6B / ~$5M)
Leveraged time series forecasting and developed branch-level performance models, optimizing operational efficiency and sales effectiveness nationwide
Developed and implemented a computer vision-based automated quality control system for fiber scanning in ultrasound liver diagnostics (FibroScan)
The system improved the accuracy and consistency of medical imaging assessments, enhancing clinical reliability and patient outcomes for liver disease diagnostics