Shawn Bakker

Parcel-level site-feasibility for Woodburn, Oregon, on real county data. It computes live in the browser and gets re-checked by a deployed geospatial API.

• The buildable-area engine unions overlapping constraints before subtracting, so nothing double-counts. It's unit-tested, and I show the browser's planar answer (51%) against the API's geodesic one (50.66%) instead of hiding the gap.
• Layered screening on real data: zoning, SSURGO soils, FEMA flood, and distance-to-sewer/hydrant utilities, plus multi-parcel assemblage with a cited yield range.
• I built an evaluation harness that caught the LLM extraction approach before production. The first real Cerebras (gpt-oss-120b) run scored 1 of 6 against a hand-verified gold set: the model returns district-level conditional rules where the product needs parcel-resolved answers. I publish the eval conditions and failure cases, not an unsupported accuracy number.

A lock-free matching engine built from scratch to survive market-open bursts, with every memory-ordering choice justified in the code.

• Lock-free SPSC queue with alignas(64) cache-line padding to kill false sharing, benchmarked at a median ~103M ops/sec. The MPMC variant uses per-slot sequence numbers with CAS retry for ABA safety.
• Price-time-priority matching with O(1) cancel and partial fills, at ~0.9 µs p50 / 2.1–2.3 µs p99 order-to-fill across 100K orders.
• Deadlock prevention by global lock ordering with a debug-mode runtime validator that compiles to zero-cost no-ops in Release. 11 of 11 tests pass, including a 50-thread deadlock torture with zero violations and a 1,000,000-item MPMC stress with zero lost.

Measured on Intel i5-12600K, Windows 11, MSVC 2022 Release, C++20, commit 1514743. SPSC throughput 93–105M ops/sec across three runs.