Polymarket

• Designed and prototyped referral fraud detection models (LR, XGBoost, Random Forest) that reduced referral fraud by 50%, by combining device fingerprints, customer behaviour features, and graph-based referral signals into a hybrid rules + ML scoring system
• Built end-to-end ETL pipelines in Python + ClickHouse + Hex to pull referral data into a unified fraud schema, with scheduled T+1 jobs and dashboards to inspect partner- and user-level behavior.

Shopify

• Deployed personalized recommendation deep learning models (HSTU, two-tower, boosting, generative) thatincreased conversion by 12% in live A/B tests through sequence-aware architectures
• Built scalable MLOps pipelines for streaming data, enabling 4× faster iteration cycles via automated training, evaluation, and microbatched inference optimizations, improving model recall by 10%
• Architected a remote inference service using Triton NVIDIA Server that isolates models from cloud deployment tools, reducing model deployment time by 60% and enabling standard technologies like vLLM while establishing the default inference deployment pattern adopted by 4+ teams.

Zomp Inc.

• Developed advanced ML models, including Random Forest, Gradient Boosting, and Convolutional Neural Networks, to enhance ILI data analysis for pipeline integrity and estimate missing pipeline properties
• Built ETL pipelines to process inspection data, improving pipeline defect detection accuracy by reducing depth measurement uncertainty by 20%, using Python

Zomp Inc.

• Developed C#/.NET applications used by 50+ analysts, which increased client productivity by 30%
• Optimized document processing and document repairs by 73% by using IAsyncEnumerable and lazy evaluation
• Applied test-driven development methodologies using the XUnit framework, achieving 95% code coverage and enabling early detection of issues, resulting in more reliable and maintainable code

Slime Scholars

• Built and maintained features for a web application with 100+ users
• Reduced database query load by 15% by developing REST APIs using HTTP protocols and refactoring data schemas in MongoDB