Built. Shipped. In production.

Insurance professionals at Faroe were querying operational data by hand: 30-minute lookups, 72-hour claims cycles, and non-technical staff locked out of their own data. While my MBA peers ran simulations for their capstones, I took the risk of a real client with real stakes.

Co-designed the system and led cross-functional technical work across data architecture, SQL optimization, API development, and conversational design, through a 27-task, 37-week engagement.

Built as a full-stack system: Next.js frontend, FastAPI backend, and LLM-assisted routing where the model selects from deterministic SQL tools rather than generating SQL, so every answer stays consistent and auditable. Multi-company data architecture with per-customer scoping, domain-based access control, multi-turn conversation memory, a reusable customer onboarding ETL pipeline, and a test framework.

Delivered to the client as a working demonstration system with a full technical handover: the foundation for their production expansion. Projected impact from our stakeholder analysis: 92% faster claims cycle (72 hours to 6), 98% reduction in data query time, and $100 to 200K in annual support savings.

Ask Remy taught me a pattern: most organizations aren't short on data, they're short on ways to ask it questions. I wanted to prove that pattern wasn't a one-time fluke tied to insurance. So I rebuilt the same architecture on a completely different domain: NASA's public Exoplanet Archive, which is free but locked behind a formal query language and a 6,240-row flat table with no visual language at all.

Designed and built the whole thing, solo and self-directed. I've wanted to work adjacent to NASA and space data for years, so I built real, demonstrable experience toward it instead of just saying I was interested.

A live pipeline, not a one-time export: Python pulls current data from NASA's API, pandas cleans it, and it lands in a normalized three-table schema wrapped in a FastAPI backend with 15 validated endpoints, each checked by hand against the raw data. The frontend frames results as a voyage: Earth at the top, real light-year distances controlling the scroll. Every planet is rendered from its actual measurements, surface color from its real temperature, light tinted by its real host star, and shown as visually unknown when the data doesn't exist rather than faked.

Live in production on Vercel: 6,240 confirmed exoplanets, explorable right now. Honest scope: it's a portfolio piece, not a research tool, and the data refresh is manual. What it proves is the transferable skill: schema design, API build and validation, and a designed, accessible frontend shipped through real deployment debugging. Visit the explorer · Backend repo · Frontend repo

EduBlck Institute needed four live research surveys and a contact pipeline without handing community data to third-party vendors, and without a monthly bill. For community-based research, data ownership is not a preference. It's the ethics.

Designed the research instruments and built both systems in code myself.

One Google Apps Script generates all four survey instruments, educator and parent tracks in short and research-partner versions, with data-integrity screening built in. A second Apps Script web app runs the contact pipeline: it validates every submission, writes it to a structured Sheet, and fires an email notification in real time. It's wired to this site's contact form right now.

Four live research surveys collecting data today, a working lead pipeline, zero vendor lock-in, and full ownership of every layer. Take a survey on the Research page

Most "AI built" websites are one prompt and a template. No architecture, no reasoning, no proof that thought went in.

Directed the full build like a design team of one: I set the story, made every decision, and used AI as the fabrication layer, not the architect.

Sitemap first. Then wireframes with click-by-click interaction notes. Then a content inventory auditing every block. Then a color system derived from color-theory research. Then a live design system. Every decision cited to published research before a single production line was written.

A repeatable, documented methodology, not a one-off prompt. The site you're standing in is the deliverable and the case study at once.