Software engineering, from firstcommit to production.

Review on nearly every pull request to the main platform repository, and a service that turns bug reports into fixes the team reviews before merge.

Created the server AI tools use to reach company data, with OAuth sign-in, write actions that need a person's confirmation and are safe to retry, and usage telemetry.

Device sign-in with a consent step, per-key rate limits, per-user write permissions, and every action attributed in the audit log.

Separate servers per domain, each with its own database roles: an HR server with tiered access that never exposes Social Security numbers, bank details or birth dates, and a 56-tool server over a partner company's golf, food-service and real-estate data.

Skiers sign up, accept a rights release before their first upload, and submit edits and photos for features or pass credit. Staff moderate every submission, and the resort ends up owning the content it uses.

Employee surveys, anonymous feedback with safety escalation, company bulletins and moderation queues, built into the operations dashboard.

A per-user daily spend cap, confirmation before any write, and a cached tool catalog, added to the team's in-dashboard assistant.

A Rust service that tracks each AI agent's spend against its daily budget, along with runner capacity and audits.

An opt-in program that matches season passes to staff records, with challenges, leagues, a reward store and a console for HR. The pilot opened at one resort in September 2026; time-off nudges are built and switch on at launch.

Typed responses, a consistent error format, request IDs, rate limiting and a convention check in CI across the endpoints AI tools call.

Phone photo in, width and length out, with a check-by-hand prompt instead of a guess near a size boundary.

Checked, checksummed exports from Sage 100 and a plain-English query layer, in internal testing, whose answers cite their source.

The company's factory projection rebuilt in code, reproducing its spreadsheet cell for cell, then fed from ERP data, with a month-by-month replay since 2021 that reproduces actual U.S. stock-outs. Container packing and draft purchase orders that a person approves come next.

Mapped where AI pays for itself across the business, from customer service to accounting data, scoped each workflow, and ranked them by payoff.

Rate card, bid sheet, job sheet and change orders as linked tabs, with a billing tracker for change orders and a flag when rates drift.

Helical and micropile install logs and load-test records built around the IBC, ASTM and FHWA requirements, with automatic pass and fail flags and an inspector countersign.

Solves for the hourly rate that keeps take-home pay flat while funding retirement, calibrated from real pay stubs, with the SECURE 2.0 catch-up limits.

A per-cycle billing checklist that catches missed trucking charges, and fleet and shop-stock checks before plow season.

A FastAPI service that reads vendor order confirmations from the inbox, PDFs included, into a de-duplicated Google Sheet with lead times and monthly spend. Built and tested; not yet switched on for daily use.

A Next.js and Supabase ordering site, ready for launch.

A setup guide for actual cost per job code in QuickBooks, with actuals pulled into the contractor's Excel budget for budget-against-actual reporting.

Clients, projects, hours, invoicing and expenses in one system, with a web dashboard and a private iPhone app I use every day.

Competing AI models each submit a weekly trade list for a small real-money account, scored against a rule-based control. Each runs under its own database role, and server-side risk limits reject any order that breaks the rules.

A checklist audit I run before every launch, covering privacy, consent, data deletion, accessibility and app-store rules, with a written report kept in each repository.

Mountain terrain, particle formations and an ark modelled in Blender, all rendering live in your browser with three.js and pausing off screen.