Plantd is turning the world's fastest-growing grasses into carbon-negative building materials. Their structural panels replace wooden panels and outperform wood on cost, carbon, and performance. D.R. Horton, the largest homebuilder in America, signed a multi-year contract for 10 million Plantd panels. That's 90,000 American homes, 165,000 metric tons of CO2 sequestered, and 1.2 million trees saved from the mill.

But the panels are only half the story. Nobody sells a machine that converts grass into construction panels, so Plantd had to invent one: a first-of-its-kind, fully electric production line in North Carolina. When you invent a new machine, there is no playbook for running it. No service manual, no industry baseline for what normal looks like. The only way to learn how the machine behaves is to watch every sensor on it, every time it runs, and remember everything you saw.

Plantd is hiring people to help build the future, from farm to factory. [See their open roles]

‍

When production tempo breaks the spreadsheet

Through development, Plantd ran the way most hardware teams do: a homebuilt pipeline of CSV exports, in-house post-processing scripts, and a quality tracking system living in spreadsheets. Machine telemetry landed in one place, operator notes in another, and production QC data in a third. This manual process worked while the team was proving the technology. It started to strain the moment they shifted from testing to manufacturing, with an operational tempo generating more telemetry than they could process.

The hardest problem was correlation. The production line produces time-series telemetry, but quality lives at the level of the individual board, and board data carries no timestamps. If a panel was out of spec, it needed to be traced back to what the press and the forming line were doing when that exact panel was made. This required custom scripts and manual reconciliation. The team even maintained a synthetic "runtime" channel in post-processing to paper over the gaps that machine stops left in the data. Comparing one production run against another to figure out "if today's batch is consistent with yesterday's" was the single biggest bottleneck in the workflow.

‍

One source of truth for the whole line

To get rid of the bottleneck, Plantd moved the entire loop onto Sift. PLC telemetry streams off the line over MQTT and lands in one platform, alongside operator annotations and panel-level results, so machine behavior and product quality finally live on the same timeline.

Instead of manually reconciling data, data review now runs automatically. Watching sensor behavior like press pressures, resin dosing, or mat thickness with Rules, which flag any run where a value crosses a threshold.

Instead of writing manufacturing observations on a sheet, Plantd engineers use Sift Annotations to mark the data itself and share exact context with colleagues.

Every new production run is automatically measured against what normal looks like with calculated channels and run-to-run comparisons.

Huade Tan came to manufacturing from SpaceX, where he spent years on console for Dragon missions. Now he wants the same discipline on the factory floor.