ArchivePulse

Inspiration

A museum conservator comes in one morning and sees that the humidity inside a manuscript display case spiked overnight. They know when it happened, but they still have to figure out what to fix. Is the case leaking? Does the silica gel need servicing? Or is the vent that lets air reach the gel blocked?

We built ArchivePulse around that moment. It helps conservators decide what to check next, compare repairs, and see whether the repair they chose actually changed how the case responds.

What it does

ArchivePulse opens in a 3D conservation workshop. Pick a display case and you can see how its humidity changed alongside the humidity in the gallery. A physics-based model ranks three possible causes based on how well they fit the readings: a blocked vent, weakened silica gel, and a leaky seal.

In our simulated incident, a blocked vent and weakened gel produce almost the same humidity curve beside the manuscript. ArchivePulse suggests checking the air next to the gel tray. Click the tray in the 3D case and you’ll see that the humidity there stayed fairly steady while the display chamber got wetter. That points toward a blocked vent. At the workbench, you can see how the case is predicted to respond to clearing the vent or servicing the gel under the same gallery conditions.

Once you choose a repair, you can save the evidence and lock in the prediction. When you have another sensor recording, you can compare it with the prediction and the original readings. ArchivePulse also helps you find a temporary case for the object during repairs and compare opening times for routine maintenance.

How we built it

We built the 3D workshop with React, TypeScript, and Three.js. The backend uses Python and FastAPI, with SQLite storing case settings, sensor recordings, and incident history. The model simulates how humidity changes across the gallery, display chamber, and gel compartment. It compares the simulated curves with the readings and uses the same gallery conditions for each repair so you can compare their effects.

For the physical prototype, we designed a miniature display case with a 3D-printed gel tray and a vent shutter that slides open and closed. The ESP32 firmware supports synchronized temperature and humidity readings from the room, the display chamber, and the gel tray. The app labels live readings, simulated incidents, and model predictions throughout the demo.

Challenges and what we learned

The hardest part was figuring out what we could tell from a single sensor. In the simulation, the reading beside the manuscript showed that something was wrong, but we needed the reading beside the gel tray to tell the two most likely causes apart.

We ran into a similar problem with before-and-after comparisons. If the room was drier after a repair, the case might look better partly because the conditions around it changed. ArchivePulse flags a comparison as inconclusive when the starting conditions or gallery humidity differ too much, or when there are too few sensor readings to judge the result.

What’s next

We want to run repeatable trials with the vent open and blocked, then compare the physical sensor readings with predictions we save before each trial. We’d also like to put ArchivePulse in front of conservators and see how it fits their work using their own cases and humidity targets.