AI engineer / Dan Gilles
From an ambiguous brief to a working pilot.
I work directly with the people who use the software. I also build agent infrastructure and contribute as a code owner to ZeroClaw, an open-source Rust agent runtime.
01 / Selected work
Agent tooling, full-stack applications, and iOS experiments. Each project states what was checked and what remains unverified.
A local control room for AI agents.
Coordinate agents across ZeroClaw, OpenClaw, Hermes, PicoClaw, and Codex. A Go backend connects to a React dashboard over SSE and WebSockets, with SQLite for storage. Runs locally, with no public demo.
Build, go test, go vet, and 7 web tests pass.
An AI customer support desk.
A customer support prototype built with Next.js 15, Supabase, and OpenAI, with pgvector for document search. The public link shows sign-in; authenticated support workflows need separate verification.
Production build passes. Authenticated workflows are not verified.
A 3D history game set at Lexington, April 1775.
A static three.js game about the night ride and dawn muster at Lexington in April 1775. It runs in the browser with no runtime dependencies, account, or API.
Build passes, 352 test checks pass, and 24 local pages load.
Recommends AI models for a task.
A standalone Go and React app that suggests which models fit a task and a priority. It works with any OpenAI-compatible provider, with OpenRouter as the default, and runs locally.
Build and tests pass. Code is local only.
Video tools for Codex. A credited port of an MIT-licensed project.
A Codex video tool server using MCP, TypeScript, and ffmpeg.
108 tests pass.
Reusable workflows for Codex.
A collection of Codex skills written in Python and Markdown, listed on skills.sh.
Skill install and scripted workflows checked; some native tests still fail.
Slack-like chat with AI replies.
A chat application built with React, Supabase, FastAPI, and pgvector. Code only; no public demo is linked.
Frontend: 55 of 55 tests pass.
Learn Chinese vocabulary through the camera.
An iOS app for learning Chinese vocabulary from camera input, built with SwiftUI, ARKit, and Core ML YOLO.
Simulator build not verified.
02 / Open source
At ZeroClaw, I own code across runtime, tools, gateway, configuration, memory, and cost. These upstream pull requests are merged.
When ZeroCode's notification stream overflowed, recovery cancelled running turns, including background sessions. The fix reattaches to the client's own running turns, keeps displayed output, and reloads the saved transcript again when the turn ends. Reconnect and resume paths retain their existing cancellation behavior.
Follow-up messages sent while an agent was using tools waited until its final answer, and cancellation could lose them. The fix pauses the tool loop when steering is waiting and delivers the message at the next step within the same turn. A long tool call still runs to completion before the agent reads the message.
A finishing turn could acknowledge a follow-up message and then exit without reading it. The fix closes the steering channel before the turn finishes and checks once more for messages already accepted. Later messages receive a closed-channel response so the client can send them as ordinary prompts.
03 / Experience
Apr 2025 to Oct 2026
Superbuilders
Built full-stack pilots and agentic workflows directly with curriculum and product teams. Turned needs into requirements and explained cost, reliability, and readiness tradeoffs. These projects were demoed or piloted, not shipped to production.
Partner: Bill of Rights Institute
Mar 2026 to present
ZeroClaw
Contribute to an open-source Rust agent runtime, with code ownership across runtime, tools, gateway, configuration, memory, and cost code.
Dec 2024 to Mar 2025
Gauntlet AI, Cohort 1
Completed a 12-week, 1,000-hour AI-first program, building a new application from scratch every week.
Nov 2023 to May 2024
University of Chicago Knowledge Lab
Created technical visualizations of neural networks and training dynamics for a forthcoming deep learning textbook.
2012 to 2021
Beijing
Taught English and STEM, and worked as a Chinese-English translator and editor.
04 / Workflow
Coding agents write much of my code. I specify the problem, review every change, test it live, and decide what ships.
05 / About
I spent about a decade living and teaching in Beijing before studying computational social science at the University of Chicago. English is my native language, and I'm professionally fluent in Mandarin Chinese.
06 / Contact
Let's talk about AI agents, developer tools, or building software with your team.
dangilles@gmail.com
