Selected work
Real systems, real capital, real users, across quantitative finance, fraud, gaming, healthcare, non-profits, and public-sector data. A sample of the work behind The Infinity Fund.
Deployed & in production
Engagements where models had to perform on real money and real users, not just in a backtest.
Designed and deployed a machine-learning muni-bond strategy on $100M of allocated capital, partnered directly with the PM and department leadership to present results and take it into production. Also supported ML underwriting and credit modeling of SoFi & Lending Club loans packaged into mutual funds.
Deployed production ML for fraud detection on fiat currency on-ramps. Built an end-to-end pipeline with batch & realtime feature engineering, SageMaker gradient-boosted training, hyperparameter tuning, and AUCPR-driven evaluation, with live model endpoints and hardened Airflow training workflows.
Optimized Candy Crush advertising bid-ask spreads using multi-armed bandits. Maintained Python bandit services, versioned Docker images to Artifact Registry, deployed to autoscaling GCP Cloud Run, and analyzed daily performance in BigQuery by geographic player distribution.
A deterministic verification engine for slot-game mathematics. Extracts a game's rules into a typed spec, then verifies paytables, paylines, RTP, and bonus mechanics against the implementation, catching math and configuration errors before they ship to the floor.
A dynamic systematic options-trading strategy on SPXW (S&P 500 weekly options), engineered from signal research through risk-managed execution. Proprietary R&D built and operated in-house at The Infinity Fund.
Non-profit & community
Pro-bono and community engagements where data and automation serve healthcare research and local non-profits.
Research with Renown Health and the University of Nevada, Reno School of Medicine: a Rust application that screens free-text clinical notes for indications of cancer. Reads messy, non-uniform provider text, runs concurrent async LLM analysis, and writes structured results, designed to extend cleanly from GPT endpoints to cheaper open clinical language models.
Data-driven automation for The Jennifer M. O'Neal Community Ice Arena. A CP-SAT optimizer builds the adult-league season schedule from available ice slots, auto-relaxing constraints when they're infeasible, while a separate engagement pipeline emails members about upcoming programs with one-click sign-up and attributes the resulting paid revenue, proving ROI for a community non-profit.
Public sector
Turning fragmented government records into structured, analyzable signal.
As a biostatistician with the Nevada Office of Analytics, built a pipeline that ingests every Nevada child-fatality PDF and normalizes it into a consistent per-county CSV schema, turning fragmented government records into analyzable data, including prior-history call counts per child.
Other projects & consulting
Engineering and research we've built, from HFT market-data lakehouses to production ML pipelines and systems infrastructure.
A Rust-based OPRA PCAP ingestion & decoding pipeline that takes options market data from raw bytes on the wire to queryable research tables, a full local lakehouse over Trino, Iceberg, MinIO, Postgres, Airflow, Superset, and ClickHouse.
A code-only, production end-to-end ML pipeline with full MLOps on AWS: Terraform-managed VPC/ECR/ECS/Fargate/S3/IAM, GitHub Actions CI/CD via AWS OIDC, SageMaker training & hosting, and Airflow/MWAA orchestration with model-registry auto-promotion.
A Rust reverse-proxy with pluggable load-balancing strategies (round-robin, least-connections) that automatically fails over to least-connections when round-robin p99 latency crosses 1s, built to internalize how a real load balancer is shaped.
Have a problem worth a case study of its own?
Start a project