Writing about building products with AI, shipping full-stack software, and the craft of product & project management.
Susan Street Fine Art's site had 150 images and very little for a visitor to do. Most of the opportunity was content and design — and a POC that showed what 13× lighter performance could look like without leaving Artlogic.
Program Edge grew from brand research to a live five-page site in two sessions — with a homepage that embeds the client build live, rather than showing a screenshot of it.
For Center X Equine Therapy, the invisible stack — Givebutter, Fillout, Airtable, Cloudflare, and a thoughtful HIPAA boundary — mattered just as much as the HTML.
The clients loved the demo — and most of the work that made that moment possible happened before I opened a code editor: competitor research, a discovery agenda, and the quiet trust patterns nonprofits need.
We're engineering bigger context windows, better RAG, smarter memory. Meanwhile the human — the one holding all the threads — is losing context faster than any model.
After a recent two-bug incident with a third-party data feed, I rewrote the rules I follow when integrating any external data source. Six items, two pieces of supporting infrastructure, and a clearer line between 'is the API up?' and 'is the data right?'
My market-risk agent told me the Strait of Hormuz was calm during a real, active crisis. Tracking down why turned into a same-day case study on the difference between an integration that runs and an integration you can trust.
I started with a familiar resident frustration: HOAs feel restrictive and hard to navigate. The useful product idea was not generic AI Q&A. It was California procedure, deadlines, checklists, and calm self-help.
Real projects built with AI agents — each one an experiment in shipping software, learning a stack, and solving problems that matter to me.
A nonprofit equine therapy website for a 501(c)(3) serving children with disabilities — warm, mobile-first, accessibility-checked, with real therapist bios, program pages, and a fundraising-ready donation flow.
Real client work for a healthcare-adjacent nonprofit — research-first discovery, HIPAA-aware form boundaries, and a tool stack (Givebutter, Fillout, Airtable) the organization can actually operate without calling a developer.
A consulting and build studio site — five pages, research-driven brand identity, live client iframe on the homepage, RSS-powered blog feed, and Web3Forms contact with budget qualification.
A professional home for client work and AI-assisted builds — not a personal GitHub profile. The homepage embeds the live Center X site to prove "I actually ship" instead of describing it.
A proof-of-concept gallery redesign for an inquiry-driven Solana Beach art gallery — exhibition-first IA, Inquire CTAs, consulting offer, and optimized WebP images at ~13× lighter than the live site.
To show that 80% content and design, 20% platform — the live site runs the same legacy Artlogic engine as blue-chip galleries, but undersells a decades-long presence with a carousel-only homepage and no conversion path.
A searchable knowledge base for learning Cursor and AI-assisted work — from beginner installation guides to deep technical articles on LLM architectures, agentic AI, and context engineering, backed by 100+ scholarly sources.
To create a free, open resource that serves three audiences at once: non-technical people trying Cursor for the first time, intermediate users leveling up their workflow, and engineers who want research papers and architectural depth — all in one site with persona-aware content that adapts to the reader.
A personal blog for my mom — a space where she can collect and share her thoughts, memories, and photographs for later generations.
To create something meaningful for family, and to learn a different tech stack — pure HTML/CSS with no framework, neural text-to-speech for audio narration, and image processing for scanned photographs from the 1950s.
A financial markets aggregator and risk assessment tool that pulls live data, scores market risk across 80+ assets, and generates a static HTML report — readable by anyone, not just traders.
To learn how to be an AI builder end-to-end: data ingestion from multiple APIs, a rule-based risk engine, Claude-powered narrative analysis, and automated deployment — all shipping to a live public URL.