Chemica II: identity, integration, and one shared Page type
How Akashi unifies five disparate scientific data sources behind one lazy-fetching Page type in D, and why picking a single identity anchor early matters more than any clever parser.
Chemica I: a GTK-based PubChem wiki app with 3D molecular viewing
GTK4 application architecture in D, a custom Cairo-based 3D molecular viewer, PubChem compound search, article rendering with collapsible panels, and dosage aggregation with multi-dimensional similarity matching.
Modernizing Selenium Support in D
WebKit GTK is abandoned on Windows, so I needed a cross-platform browser engine for Insurgent. D had no working Selenium bindings, so I forked a 2018 JsonWire library and brought it up to W3C WebDriver.
Selenium II: dispatch-oriented browser automation
Replacing fragile linear click-and-sleep scripts with an observable state machine and event dispatch queue. Stale-element errors went from constant to almost never.
Selenium I: automating multi-page workflows with Ruby and Selenium
Building a robust Ruby automation from 500 to 700 lines with explicit waits, DOM-driven state machines, CDP hooks, layered retry logic, and modular page objects.
Design that matters: what a Selenium SDK taught me about my AI SDK
Carrying the same layered-contract thinking from the Selenium SDK into an AI SDK rework, and what changed when the layers had to share nothing but a contract.
Semantic retrieval on laptops: embeddings, centroids, and SIMD
Running embedding-based search on consumer hardware with two-stage centroid pruning, AVX-optimized dot products, and int8 quantization to keep the working set in cache instead of on disk.
Hacking C#: dynamic syscall stubs and user-mode hook detection
An educational walkthrough of syscall-ID extraction, shellcode stub generation, and hook taxonomy in C# on Windows, built on top of CoreCLR internal structures.