Guided tour
See the lab in three workflows
Three short, silent screen recordings walk through the main journeys: parsing a messy lab sheet, catching a shift on a control chart, and reviewing the runs that clustering puts outside every group. Every step is captioned on screen and listed beside its video; select a step to jump to it. About 3 minutes in all.
Everything shown uses synthetic data and the default data seed (30034), so you can repeat each journey exactly. The recordings come from a Playwright script that also runs as an end-to-end test of these journeys. No API key was used: the two AI screenshots show a mocked response and say so.
Tame the spreadsheet1:02
Load the synthetic messy lab sheet and step through the parser: the rule cascade, what it extracts from free text, and where the rows it cannot read go.
Steps and transcriptThe captions shown in the recording. Select a step to jump to it.
Spot the drift0:49
Open the control charts, inject a sustained shift, watch the run rules fire, then read the false-alarm against detection trade-off from the simulation study.
Steps and transcriptThe captions shown in the recording. Select a step to jump to it.
Find the odd runs0:47
Project the synthetic peak measurements, cluster them, tune by silhouette, and review the injections the clustering puts outside every group.
Steps and transcriptThe captions shown in the recording. Select a step to jump to it.
Key features
Screenshots at 1440 × 900 and on a 390 px phone, in light and dark mode. Select one to enlarge it.
On a phone (390 px wide)
How these were made
A Playwright script drives the installed Chrome through each journey at 1280 × 800, adds the caption banner and a visible cursor, and asserts what every step should show, so a broken journey fails the run instead of producing a misleading video. ffmpeg turns the recordings into these MP4s and the README's GIFs, and the caption timings into the step timestamps and WebVTT captions. The AI shots intercept the request to the provider inside the browser and answer it with a reply written by the script; see the AI use statement for what the real feature sends.