Building OpenCastor
Declarative robotics
starts with
a manifest.
ROBOT.md is to a robot what CLAUDE.md is to a codebase.
One file — YAML frontmatter + markdown prose — so Claude Code, Claude Desktop, Cursor, Zed, Gemini CLI, or any MCP-aware
agent can safely operate the robot. I'm a one-person team building the spec, the runtimes, the surfaces, and the public
registry — almost entirely with Claude Code.
Try it on your robot
pip install robot-md robot-md-mcp
robot-md init --yes
The Ecosystem
Eleven repos. One stack.
A protocol at the bottom (RCAN), a manifest in the middle (robot-md), MCP and Agent SDK surfaces on top, a public registry on the side. Each layer is its own repo, its own release cadence, its own clear concern.
Method
Built with Claude Code
The point isn't that an LLM helped me write code. The point is that a one-person team — me — was able to ship and maintain a multi-repo ecosystem because Claude Code's long-context behavior made the cascade tractable.
A spec change in rcan-spec ripples through
rcan-py and
rcan-ts; a manifest schema change ripples through
robot-md, every consumer, and the registry. Claude Code holds
the context that would otherwise be split across a team.
The Anthropic-native bias of the surfaces — MCP server, Claude Code plugin marketplace, Agent SDK on the pendant and dispatcher, Claude as the exclusive eval model in autoresearch — is on purpose. Every Anthropic primitive gets a first-class surface.
By the numbers
April 2026 · single-person team
Hardware in the loop: a SO-ARM101 arm called bob, registered as RRN-000000000001.
Every shipped feature is exercised against real servos before merge.
Recent activity
What I'm shipping now
The projects with the freshest commits and posts — surfaced automatically from the build log.
Pantry Atlas
Introductory post coming soon.
Latest · May 28, 2026 Shipping a 227 MB database and a 1.9 GB image to the public for $0/month →Scientific Skills & Visualization
A toolkit of Claude Code skills for scientific work — Tufte-grade chart critique, 3D molecular rendering, and bundler-free scrollytelling.
Latest · May 27, 2026 When a chart-critique skill catches a design bug before ship →Heat Protein & Metrics Lab
Browser-based scientific tooling for occupational heat safety — WBGT metrics, heat-shock protein visualization, and molecular-dynamics in the browser.
Latest · May 27, 2026 Building Chapter 7 — a sandbox and a live NWS panel that share one set of formulas →Heat Threshold
Built at the Cerebral Valley × Google DeepMind I/O hackathon — a heat-safety dashboard with a WebXR spatial HUD and a Gemini Live ↔ Managed Agents voice bridge, all shipped in a seven-hour window.
Latest · May 24, 2026 Heat Threshold at the Google I/O hackathon — overbuilt a lot of things →Earlier work
Before robot-md
Computer vision, accessibility, and edge AI projects that informed the choices in OpenCastor.
LiveCaptionsXR
Spatial captions for XR collaboration. On-device speech recognition with anchored transcripts so deaf/HoH participants can follow conversations inside headsets.
Project page →BicycleRadar
Predictive collision avoidance for cyclists. Sensor fusion + ML produced 3.2-second warning windows at 96% prediction accuracy.
Project page →Dronevade
Edge computer-vision platform for drone detection. Custom YOLO models, RF + thermal fusion, designed for wildfire responders and utilities.
Project page →HeatCompass
Personal heat-stress monitoring for outdoor workers and athletes. Edge inference on a wrist-worn device.
Project page →Latest writing
From the build log
Shipping a 227 MB database and a 1.9 GB image to the public for $0/month
How PantryAtlas distributes large artifacts on Cloudflare R2 with zero egress fees — the real cost math, why download-cost 'abuse' is bounded by design, the one setting that actually matters, and the gotchas (wrangler's 300 MiB cap, the 512 MB cache ceiling) I hit getting there.
Ranking 50,000 recipes on a CPU that embeds 16 phrases a second
The engineering behind PantryAtlas v0.2.0: an instant→refine→swaps architecture forced by the Raspberry Pi's embedding budget, a ranking formula that only ever settles downward, relaxed-JSON Gemma over grammar-constrained decoding, and offline-first sync that replays on reconnect.
Introducing PantryAtlas — a recipe brain that lives in your kitchen, not the cloud
PantryAtlas is local-first culinary AI for a Raspberry Pi 5: tell it what's in your pantry, get ranked recipes back — 49,965 of them, matched on-device, fully offline, no account, no subscription. Here's what it is and how to get it.