Portfolio

Led technical architecture for a platform that automatically designs fixed-wing drones by selecting optimal components and tuning airframe geometry. Custom genetic algorithm with KD-tree component selection; complete AWS deployment (150+ Terraform resources); coordinated 5–15 engineers across 6+ repositories.

Transformed a 3,000-line AI-generated notebook into a production ML pipeline estimating ML expert headcount across 400+ organizations from 250,000+ LinkedIn profiles. Developed a novel multivariate probit bootstrap model with correlated annotators and five-layer uncertainty quantification.

Offline-first PWA for waste management teams at large events such as music festivals. React/TypeScript with InstantDB real-time sync, interactive Leaflet maps, automated Google Slides/Sheets reports, and Cloudflare Workers/R2 infrastructure managed with Terraform and SOPS.

Mathematical framework for allocating advertising budgets across 18+ channels to maximize campaign reach. Poisson process model for media engagement, Bayesian framework for imperfect demographic targeting, formulated as constrained nonlinear optimization. Documented in a 14-page mathematical specification derived from first principles.

3D functional-structural plant modeling for a 30-meter solar greenhouse with 4,032 plants. Monte Carlo ray tracing (1M+ rays) for light simulation, Bayesian optimization over 8 architectural parameters, 6-point Legendre-Gauss quadrature for daily averaging — achieving 29% improvement in photosynthesis efficiency.

Publication-quality data visualization for a connectomics and brain simulation research team. Multi-scale figures spanning many orders of magnitude, innovative 3D visualizations, and custom heatmaps for neural recording capabilities across four iterative phases through final publication-ready deliverables.

High-performance portfolio rebalancing library in Rust using the Clarabel.rs SOCP solver. Extended Markowitz mean-variance optimization with novel soft constraint formulations; Python bindings via PyO3. 50× performance improvement over the Python baseline (5 ms → 100 μs per solve).

Contributed to Kraken's first IRS-compliant tax reporting microservice. Rust/Actix backend with PGP encryption for vendor data exchange, Prometheus/Grafana monitoring, and Terraform-managed infrastructure. Served ~4,000 users in the first weeks of deployment.

Python pipeline to interpolate 2GB+ irregular seafloor depth datasets into 10m-resolution GeoTIFFs. Resolved a 769GB memory bottleneck by switching to cubic interpolation; implemented block-average decimation, trend removal, and convex-hull masking for correct geospatial alignment.

Implemented a bipartite matching algorithm from a research paper to find row/column permutations that maximize the product of matrix diagonal elements. Handles extreme numerical ranges (10⁻²³ to 10¹⁴). Validated against MATLAB's equilibrate function; client awarded a bonus for the delivery.