Builds

The four-document pipeline I use to run long AI-assisted engineering efforts — research, plan, execute, release — so a dozen memoryless chats add up to one coherent build.

A sync system that mirrors an AI coding assistant's skills, hooks, and settings across machines via a private Git repo, gating anything that auto-executes behind explicit human review.

Built a local-first sync system across two machines so my second brain and its AI assistant are always current and reachable from my phone, with no cloud host in the loop.

A weekend self-challenge: take a domain I knew nothing about (prediction markets) from zero through full discovery, architecture, and a working multi-LLM research dashboard.

Built a vendor-to-system fee matching pipeline in 5 hours that automated a manual billing reconciliation process, pinpointing genuine discrepancies across 13,000+ matched packages per billing period.

Built a 5-dashboard operational metrics suite in 9 hours across four legacy recording systems, establishing the quantitative baseline that new-platform performance will be measured against.

A Claude Code skill paired with a static portfolio site that turns project documentation into a 5-minute guided conversation, so accomplishments are recorded while context is still fresh.

Implemented greenfield cross-system analytics across two repositories, creating the organization's first workflow-grain measurement of end-to-end document recording time.

A Claude Code skill that translates plain-English test intents into API calls and browser interactions, cutting analytics event verification from 10 minutes to 10 seconds.

A suite of Claude Code skills that encode platform tribal knowledge and auto-generate complete document recording integrations, compressing a month-long onboarding process into a single session.

Built an automated Python pipeline that measures AI document extraction accuracy against ground truth at scale, replacing 8 days of manual work per batch with a 15-minute command.

Proved out AI-powered text extraction as a replacement for manual document review, integrating a third-party extraction service and validating that its output was accurate enough to feed directly into the product's website.

A personal proof-of-concept testing how far fully autonomous AI coding could go with zero cloud dependency — a plan-write-test-commit agent loop running entirely on local hardware via Ollama.

A PM workspace where pressing one button turns a precisely-specified PRD into a working feature branch — Claude plans, implements, tests, and commits every requirement autonomously.

A self-imposed crash course in native iOS: build a working SwiftUI/SwiftData app from zero, using a multimodal meal-logging tool as the vehicle rather than the goal.

A live-streaming gamma/vanna/charm exposure dashboard with full historical replay, built to replace paid delayed-snapshot options tools.

A unified sector rotation, macro, and news dashboard with regime scoring and a stock scanner, replacing a multi-tool morning routine.

A frictionless daily/weekly trading journal with reusable templates and an AI analysis layer, built into a personal trading toolkit.