We build like a team.
There are two of us.
Everything we ship at FC compounds — the research feeds the strategy, the strategy ships as tools, the tools get pressure-tested and deployed. Each layer makes the next one faster.
How We Got Here
This wasn't planned. It emerged from a simple realization: every session with Claude Code could ship something real. Here's what happened.
One HTML file. A tool to help international student ministry workers understand their context. No framework, no build step — just a single file deployed to Cloudflare Pages in 20 minutes. It worked.
Every session can ship. The pattern became clear: identify a real problem, build a focused tool, deploy it the same day. One session, one tool. The portfolio grew faster than expected.
Quality became the constraint. Fast shipping is easy. Shipping things that don't embarrass you is harder. We built the roast cycle — 3 rounds of 10-expert panel review, 150 findings per project — to pressure-test AI-generated work before it goes live. Speed stayed. Slop went away.
The Missions Library changed the scale. We scraped 161,000 articles from 101 missions and faith-and-work organizations. That corpus didn't just become searchable — it became the research base for the Faith & Work Hub strategy, partner analysis, and content tools. Research and product stopped being separate tracks.
50+ deployed projects. A custom skill library. Automated review loops. Research infrastructure that feeds live products. And a workflow that gets faster the more you use it — because every session builds on what came before.
Why It Gets Faster
Each layer of work feeds the next. This isn't a collection of projects — it's a system.
101 sources
Missions Library
9 documents
$800K plan
for field use
real users
Vercel · Workers
live in minutes
The corpus isn't just data — it surfaces what practitioners actually talk about. When we built the Faith & Work Hub strategy, we weren't guessing at what Asian returnee professionals need. We had 161,000 data points from the organizations already serving them. The strategy came out of the research, not despite it.
The strategy isn't just a plan — it becomes the brief for tools. Clear audience, clear problem, clear success criteria. That's why tools built from the corpus take a session instead of a week. Half the thinking is already done.
The tools go through the quality loop so they don't embarrass the org. Three rounds of expert review, a 30-point anti-AI audit, and real deployment. Not perfect — but intentional.
And because the infrastructure already exists — the CLAUDE.md files, the custom skills, the deploy pipeline — the next thing is always faster than the last one.
What We Reach For, and Why
No version numbers, no install instructions. Just the tools and the reason each one earned its place.
/roast /deploy /resume /wrapup — the parts of the workflow that happen every time, automated. Each skill is a slash command that launches a specialized agent loop.How a Build Actually Works
Walk through a real build from start to finish. This is the CAMH Directory — 550+ verified Asian American Christian therapists, built in a single session.
Start With the Problem
550+ Asian American Christian therapists exist, but they're scattered across Psychology Today, PAAC, and personal websites with no centralized, searchable directory. A real gap. A fixable one.
Scrape the Data
Python scripts pull from Psychology Today and PAAC. Structured CSV gets cleaned into JSON — therapist name, location, specialties, languages, insurance, photo. No AI needed here; just file I/O and parsing.
Ship the First Version
Single HTML file. Fuse.js for client-side fuzzy search. Tailwind CDN for styling. One wrangler pages deploy command. Live at camh-directory.pages.dev. Total time: one session.
Run the Roast
Three rounds of 10-expert panel review. 150 findings total. Accessibility, performance, copy clarity, trust signals, mobile layout. Prioritize the ones that matter, implement them in order.
Anti-AI Audit
A 30-point checklist that finds AI aesthetics: gradient washes, vague copy ("empowering individuals to thrive"), corporate jargon, excessive border-radius, shadow stacking. Replace with specific language and human-feeling design. This is what separates a tool people trust from a tool that feels generated.
Deploy and Move On
381 therapist photos matched and loaded. State pages added. Submit form wired. OG image generated. Sitemap built. Done — and the loop is exactly the same whether it's a 1-page tool or a 10-page product.
The Highlight Reel
Five projects that show range: from the org's public face to a co-built product to a client site to a real community tool. Not a complete list — the complete list has 50+ items.
- 16 pages, 198 commits, fully deployed
- Collective impact framework, scholarship program, Faith & Work Hub
- Most committed-to codebase in the portfolio
- Azure Pronunciation Assessment API for real phoneme scoring
- Gamification: XP, streaks, 18 achievement badges
- Chinese-first i18n, diagnostic, drill, spaced repetition
- Next.js 15, TypeScript, full rule engine for eligibility logic
- Multi-step quiz, results page, document vault
- The most technically complex build in the portfolio
- Scraped from Psychology Today + PAAC via Python
- 381 therapist photos, state pages, submit form
- Zero server — client-side Fuse.js search over 187KB JSON
- 7 pages: home, about, impact, apply, governance, partners, 404
- USD/JPY calculator, impact metrics, partner network
- Agency-level visual polish — Barlow Condensed, tight grid
A Running System
If you're joining FC, you're not starting from scratch. The tools are here, the methodology is documented in every CLAUDE.md, and everything built so far is open to the next person who wants to take it further.
What's coming next: the Faith & Work Hub listening tour (50 returnee interviews in Singapore), real users on Accent Trainer, Care Navigator going public, and whatever surfaces from the corpus next. The pipeline doesn't stop.
The first thing to run when you open any project:
/resume
# See the full portfolio at a glance
/portfolio
# Ship something
/deploy
The methodology is learnable. The infrastructure is already running. The only thing left is to build.