What if a coding agent didn't stop after one pass?

Superloop is my attempt at making agents behave less like one-shot assistants and more like accountable execution workers: give them a mission, a budget, a finish line, and proof that the job is actually done.

mission → round → verify → learn → repeat → stop

_{Agent harness · Python · Codex · Claude Code · developer tooling}

Analytics usually gives you dashboards. I wanted the story.

TideStat follows the path behind a dollar — from acquisition and anonymous visitor behavior to conversion and verified payment — and makes that journey visible through live maps, visitor characters, Website Footprints, and Revenue Stories.

acquisition → visitor → behavior → conversion → payment → revenue

It currently spans a browser SDK, Shopify/Stripe attribution, Cloudflare Workers + D1, real-time visualization, bilingual UI, and a slightly unreasonable amount of attention to how analytics should feel.

_{Revenue analytics · Cloudflare · D1 · Three.js · SDKs · product design}

Where unfinished ideas get shipped before I overthink them.

My personal AI lab: part portfolio, part research notebook, part playground. It's where I experiment with AI-native interfaces, programmatic 3D/WebGL, motion, publishing, creator tools, and whatever seems interesting enough to build that week.

Recent experiments include a robotic-hand WebGL hero, scroll-linked motion, D1-backed comments + swipe moderation, bilingual publishing, and Remotion-based creator workflows.

_{AI products · Next.js · WebGL · motion · Cloudflare · Remotion}

A floor plan you can walk through instead of stare at.

A ceiling-free interactive apartment model with real / white-model / wireframe modes, room walkthroughs, day/night lighting, furniture inspection, and a floor-plan view — modeled around real dimensions and clearances instead of decorative 3D alone.

_{Three.js · spatial UX · realtime 3D · interactive visualization}

• AI that acts, not just chats. I care about loops, tools, verification, memory, state, and getting to an actual finished result.
• Products with feedback loops. Build it, instrument it, watch what happens, then let reality decide the next iteration.
• Making invisible systems visible. Traffic, agent state, user journeys, spatial relationships, market signals — I like interfaces that make abstract systems feel tangible.
• Markets as systems. I keep exploring quantitative research, market structure, risk tooling, and ways to turn noisy signals into explicit decisions.

I started much closer to the metal — databases, distributed systems, Kubernetes, observability, cloud-native operators.

These days I work higher up the stack, but the obsession is basically the same: state, reliability, feedback loops, and systems that keep working after the demo ends.