Work
A mix of client work, open-source AI tooling, and writing — each shipped end to end. I look at every project through two lenses: a product lens (start with the user, find the unmet need, set a North Star) and a program lens (notes, action items, a task list, dependencies, and a critical path that gets it launched).
Featured case study · Client engagement
A mobile-first, accessible website for a 501(c)(3) equine-therapy nonprofit — helping families understand the program, trust the team, join the waitlist, and support the mission. But the site is only the visible part: I delivered the entire technology stack end to end — domain and DNS, the inquiry and volunteer forms, the donation platform, hosting, and compliance posture. For an org with no IT team, I was the IT team.
Live preview — the real, deployed site, embedded right here and currently in client review.
The challenge
The organization had a clear purpose — licensed therapists using the movement of the horse to help
children and adults build balance, strength, and confidence — but its online presence didn’t carry
it. There was a thin, basic existing site on a .com address with little for a visitor to do,
and the .org they wanted to use was a parked “Launching Soon” page. No real way to
join a waitlist, no integrated donations, no mobile-first or accessible build, and a small team with
no developer, no IT staff, and no time to wrangle domains, forms, and donation tools.
The job was to turn a trusted, in-person service into a website — and the working technology behind it —
that earns that same trust in the first five seconds.
Before & after
Same mission, same story — rebuilt as a fast, accessible site with real ways to act,
on the org’s own .org domain.
.org domain.How I approached it
Good outcomes need both: the product thinking to build the right thing, and the program discipline to actually ship it.
Product lens
Program lens
The discovery trail
Before designing a single screen, I studied how the best therapeutic-riding nonprofits earn trust online — then turned the patterns into an information architecture.
The Shea Center, Ride Above Disability, Tara’s Chance, and SD Therapeutic Horsemanship — national polish and local peers, to see what families and donors expect.
Emotional hero + one mission line; three CTAs (Donate / Volunteer / Get started); named parent quotes; an accreditation/trust signal; a “what is equine therapy?” explainer for newcomers; meet-the-horses with sponsorship; and impact numbers.
A first-launch sitemap — Home, Services (PT/OT), Meet the Team, Meet the Horses, Waitlist/Inquiry, Donate/Support, Sponsors, Contact — each section mapped to a user need and a conversion.
A polished homepage and core pages first, reviewed together, then refined — collaborative, not a hand-off.
Implemented as fast, accessible pages wired to the services that run the org; checked on a 375 px phone; deployed to a live, HTTPS URL on Cloudflare Pages.
The full stack, delivered
A website is only the part you can see. Most small and mid-size organizations don’t have an IT team to stand up a domain, wire forms to a spreadsheet, choose a donation platform, or keep email working when DNS changes. For Center X, I owned all of it — strategy through hosting — so the client never had to coordinate a registrar, a developer, a forms vendor, and a payments tool on their own.
Here’s what that meant tool by tool — the actual decisions behind each layer:
After weighing Google Forms, Airtable, Tally, and Fillout (generous free tier, unlimited forms from one account), I built the Waitlist Inquiry, Volunteer Interest, and General Contact flows — each routing to a tracked destination (Sheets/Airtable) with category and source tags. Deliberately limited to basic contact, service interest, and a short inquiry, with explicit “please don’t submit medical details” copy.
Because the public forms collect no protected health information, the marketing and fundraising site shipped now. The form is a swappable module: when Center X needs real clinical intake, it moves to a BAA-signing vendor — Jane App for a PT/OT practice, IntakeQ/PracticeQ, or Jotform HIPAA — separate from the marketing site. (Healthcare judgment carried over from FDA-cleared medical devices at Dexcom.)
After comparing PayPal, Zeffy, and Donorbox, I shipped Givebutter — now live on the site. Genuinely free on the optional-tip model (tips on = $0 platform + $0 processing; tips off = a flat 3% + standard processing), with a free donor CRM, recurring giving, and campaign notifications. An Amazon Wishlist is an easy in-kind supplement.
Deployed on Cloudflare Pages: free, a fast global CDN, automatic HTTPS, Git-based deploys with preview links for client review, and a password gate that was trivial to toggle before launch.
Kept the domain registered at GoDaddy (no transfer needed) and pointed DNS at Cloudflare
for free DNS, SSL, and CDN — with email records preserved so mail kept working through the cutover.
The site now resolves on its own .org with automatic HTTPS.
Every form lands somewhere with an owner — a shared Sheet or Airtable base per category (waitlist, volunteer, contact) with the tags built into the form structure, so no inquiry goes unanswered.
The outcome
How we worked together matters as much as what shipped: embedded with the client, sharing context, capturing every decision, and making the calls together. A senior collaborator, not a vendor.
More work
Smaller write-ups — each one shipped, public, and built to learn a stack or solve a real problem.
An Apache-2.0 framework that detects when AI agents skip tool calls or fabricate outputs — silent failures invisible to normal observability. Product lens: a category most teams can’t even name yet. Program lens: 310 tests, five framework adapters, and a live dashboard — shipped, not slideware.
View projectAn automated analysis agent covering 55+ assets with a 3-layer risk engine, built on Python, Claude, yfinance, and FRED. Product lens: make market risk readable by anyone, not just traders. Program lens: deployed daily via GitHub Actions with cross-source validation so a broken data feed can’t fail silently.
View projectField notes from real builds — how I work with AI agents, what context engineering actually takes, and the failures worth learning from. The thread: thinking out loud, in public, so the craft is visible — not just the polished result.
Read the blogWhether you need something shipped or a clear plan to get there, let’s talk about what an edge looks like for you.
Start a conversation
