Field catalog
Specimens from the deep.
A loose log of things I've shipped, prototyped, or kept pinned to the corkboard.
Sites & apps
Released into open water.
Sites and apps built for clients — some public, some login-gated.
- App:
SiteSee
Field-inspection app for HybriX Energy — observations in, reports out.
- year
- 2026
- stack
- React
- stack
- Hono
- role
- client
- pwa
- client
- react
Technicians log observations and inspections on their phone — work breakdown, photos, participants — and structured reports land in the right inboxes. React PWA + Hono API in an all-TypeScript monorepo. Login-gated, so no tour, sorry.
- App:
Ringtime
Voice-AI recruiting agents for Ringtime — they call, screen, and match candidates in 22 languages.
- year
- 2026
- role
- client
- voice-ai
- agents
- client
Blue-collar recruitment runs on phone calls, so the product is one. I joined just after the company started — founding AI engineer, building the agent infrastructure and figuring out what MLOps looks like when your product is a phone call.
- Site:
Maus Cars
Car detailing & auto-import — Kortrijk.
- year
- 2026
- stack
- Astro
- stack
- Cloudflare
- role
- client
- site
- client
- astro
One studio, two trades, so the site is a hub with two spokes: / routes visitors to /detailing or /import. Astro on Cloudflare, built to load fast on a phone in a parking lot.
- Site:
Termicolor
Drop an image, get a terminal palette — WCAG-aware, export anywhere.
- year
- 2026
- stack
- Astro
- stack
- React
- typescript
- react
- design
Picks accessible foreground/background pairs against the WCAG contrast curve and exports to iTerm, Alacritty, Warp, and a handful of other terminals — no signup, no upload, all in the browser.
Code
Tackle box.
Open-source tools and libraries — take what you need, leave a star.
-
Code:Fence
A lightweight LLM & agent library — the bloat moat, vs LangChain.
- stars
- 49
- downloads
- 2.5k/mo
- license
- MIT
- year
- 2024
- python
- llm
- open-source
Built to stay small while still composing chains, tools, and agents. Typed inputs, streaming support, and a model-agnostic interface that doesn't force you into a framework just to call an LLM twice.