Kaelbot

Django/Python · React/TypeScript · PostgreSQL + pgvector · Celery · Redis · Docker

Kaelbot is an autonomous AI job-search agent that Zak designed and built solo: a Django/Python backend and a React/TypeScript frontend, with PostgreSQL and pgvector for semantic memory, Celery and Redis for background processing, and Docker for deployment.

The agent orchestrates its own tool calls against more than twenty-five structured tools — lead creation, resume tailoring, cover-letter generation, dossier synthesis, artifact storage, preference clustering — and maintains per-user persistent memory that supersedes corrected facts over time. It runs autonomous web-research investigations, generating its own search queries and synthesizing the results into ranked leads with fit scores, and it generates tailored resumes and cover letters that re-emphasize a user's existing material without inventing experience, with factual integrity enforced at the prompt layer.

Kaelbot is not currently deployed publicly. This entry describes work that was designed, built, and operated — not a live product.

Royal Tara Cove

React Three Fiber (Three.js) · WebRTC peer-to-peer multiplayer

Royal Tara Cove is a browser-based 3D neighborhood game Zak built for his kids, Penny and Luke: a recreation of the cul-de-sac in Avery Ranch, Austin, TX, where the family used to live. It runs in React Three Fiber with real-time peer-to-peer multiplayer over WebRTC, so Zak and his kids can be in the old neighborhood together from separate browsers, choosing a character — Dad, Penny, or Luke — and moving through a street modeled house-by-house from the family's memory of the real place.

The game has grown over dozens of iterations, with modes and details added as the kids come up with ideas for what the neighborhood should have next.

Play the live game → royal-tara-cove.netlify.app