Skip to main content

AI portfolio — 2026

AI tools, built slowly. Documented honestly.

Four live products. A multi-agent system that ships real commits. An accessibility map shaping how disabled neighbours navigate the city. All open source — and built one careful deliverable at a time.

Live Projects

Built and shipped. Open source. Everything in this portfolio is live.

Four products on the open internet. Click any title to open the GitHub repo or the running app.

789

tests passing

AccessMap

  • Mobile
  • WCAG AA
  • Open source

15

AI agents

Claude Corp

  • MCP
  • Real commits

50+

features shipped

Prompt Library

  • No backend
  • Browser-only

E2E

encrypted

Mutual Mesh

  • Privacy-first
  • Community

Work

A handful of recent things, made with intention.

Solo builder · 2026

AccessMap

Mobile app for flagging accessibility barriers — broken ramps, missing tactile paths — pinned to a real city map. Privacy-first: no tracking, no data sold.

  • Expo
  • React Native
  • Supabase
  • TypeScript

Architect · 2026

Claude Corp

An 18-role AI team governed by a written Constitution. Each agent owns a domain; an orchestrator runs the cycles. The infrastructure behind this portfolio.

  • Claude Code
  • MCP
  • Multi-agent orchestration

Solo builder · 2026

Prompt Library

Local-first prompt manager in Next.js 15. Prompts stay in the browser, API key never leaves the machine. Fifty features shipped across twenty stacked branches.

  • Next.js 15
  • TypeScript
  • Vitest
  • localStorage

Solo builder · 2026

Pac-Man Code Trainer

CLI commands turned into a retro arcade game. Forty flashcards covering Claude Code and macOS shortcuts, zero dependencies, runs straight from GitHub Pages.

  • Vanilla JS
  • HTML
  • CSS
  • GitHub Pages

Lead engineer · 2026

Mutual Mesh

Mutual-aid app connecting neighbours who want to help with those who need it. Privacy-first: a PRIVACY.md gate blocks every new feature before any code ships.

  • Expo
  • React Native
  • Supabase
  • TypeScript

How I work

Three quiet steps, repeated carefully.

Discover

Start with the smallest, most honest version of the problem. Talk to the people who will live with the thing — not just the people who will buy it.

Build

One careful slice at a time. Type-safe, accessible from the first line, instrumented enough to learn from. Refuse to ship what I haven't tried to use.

Ship

Document what changed, what's still rough, and what the next maintainer will need to know. A deliverable is the work plus the story of how it was made.

About

I build AI tools with care.

One careful deliverable beats a dozen rough ones.

Most of what I make starts with a problem worth solving, then a small thing that solves it well. I work from Canada, mostly on AI-assisted tooling, accessibility, and the quiet infrastructure that makes a product feel calm to use.

I keep a written record of how each thing was built and why — both for the people who come next and for me, the next time I need to remember.

Read the full story

Certificates

Credentials earned along the way.

  • Anthropic

    AI Fluency Framework & Foundations

    May 2026

    View
  • Anthropic

    Introduction to Claude Cowork

    May 2026

    View
  • Anthropic

    Claude 101

    May 2026

    View
  • Google

    Google Prompting Essentials Specialization

    July 2025

    View
  • University of Michigan

    Programming for Everybody (Getting Started with Python)

    June 2025

    View
  • DeepLearning.AI

    AI For Everyone

    June 2025

    View

Contact

Have an AI project worth building?
Let's talk.

Reach out at skylerhalisky@gmail.com

Get in touch