Project 01 / 06
Open beta
ProPulse
HF propagation forecasting for radio operators, from an ITU-R P.533 engine and live signal reports.
At a glance
- Codebase
- More than a million lines of code
- Physics
- ITU-R P.533, validated against the ITU’s reference code
- Data
- NOAA space weather, PSK Reporter, Reverse Beacon Network, DX clusters
- Infrastructure
- Vercel with 80+ edge endpoints, collectors on Railway
- Delivery
- 1,000+ pull requests merged by the agent fleet
Overview
ProPulse is a real-time propagation platform for amateur radio operators. The tools most operators rely on report the bands as they are now, and ProPulse goes further by forecasting when a band, and a path to a particular part of the world, will open. It pairs my own implementation of the ITU-R P.533 prediction method with live data from NOAA’s space weather feeds, PSK Reporter, the Reverse Beacon Network and DX clusters, and puts the results on PropSphere, a 3D globe, beside solar conditions, a DX planning wizard, a contest engine and a logbook. It’s a mobile-first web app in open beta. The codebase is more than a million lines of code, and a fleet of AI agents built all of it under my direction.
What it does
PropSphere
A three.js globe with more than 40 layers, from the ionospheric shells to live spots, aurora, lightning and satellites.
Band forecasts
MUF, LUF, circuit reliability and the best band for any path, computed in the background across the whole map.
Solar Pulse
Live solar and geomagnetic conditions, the forces that open and close the bands.
DX planning wizard
Plans long-distance contacts against the forecast, before you key up.
Contest engine and logbook
Contest operation and logging in the same app as the forecasts.
The story
Ultraviolet and X-ray light from the Sun strips electrons from the thin gas of the upper atmosphere, and the ionized layer it leaves, the ionosphere, refracts HF signals back toward Earth, so a signal can hop across an ocean. That layer changes by the hour: daylight builds it up, darkness lets it decay, and when bursts of solar wind strike Earth’s magnetic field they can set off geomagnetic storms that close paths altogether. ProPulse models that physics and checks it against live reports from stations around the world, to show where a signal can reach now and when a band will open next. My hope is that this modeling will support forecasting for emergency communications and help operators reach one another, near and far, even in the most demanding conditions.
How it’s built
I created ProPulse, and a fleet of AI agents builds and maintains it under my direction. It began as research into solar weather and the tools radio operators already use, and that research turned up a gap. Most of those tools report the bands as they are now, and few forecast when a band, or a path to a given part of the world, will open. I planned and designed ProPulse with my agents to fill that gap, then built the GitHub tooling that lets many of them work on one codebase at once without colliding. It is now more than a million lines of code.
- A physics engine, built to the standard
- At its core is my own implementation of ITU-R P.533, the International Telecommunication Union’s method for predicting HF circuits. It models electron density along the path, received signal strength and the noise the signal has to overcome, and its output is validated against the ITU’s own reference code. From it, ProPulse computes the maximum and lowest usable frequencies, circuit reliability and the best band for any path, with map-wide forecasts running in the background.
- Every hop, traced
- For each hop of a multi-hop path, the engine finds where the signal lands and the height at which the ionosphere turns it, then accounts for what it loses to D-layer absorption and the aurora. Where the signal comes back to Earth, a 3D model of the ground surface works out how land, sea and coastline reflect and scatter it.
- The whole station, modeled
- ProPulse models the station itself, physically. It follows the feed line from the radio through every cable and connector, the amplifiers, switches and chokes, out to production-grade antenna systems on masts as tall as 150 feet, so a forecast accounts for the equipment that will carry the signal.
- PropSphere, a globe of my own
- The globe is built on three.js with custom shaders and carries more than 40 layers, from the shells of the ionosphere to live signal reports, the aurora, lightning and satellites.
- Live data, around the clock
- Collectors on Railway ingest PSK Reporter, the Reverse Beacon Network, DX clusters and NOAA’s space weather feeds day and night. The app runs on Vercel and serves more than 80 edge endpoints.
- Built by a fleet of agents
- Agents running on Claude, Codex, Cursor, Gemini and Grok take issues from one shared board, each in its own checkout of the repository. A GitHub review bot I built checks every pull request, and a merge agent sequences the approved ones, resolves conflicts and merges them one at a time. More than a thousand pull requests have landed this way.