ClearLine

ClearLine is AI line clearance for pharmaceutical packaging lines. I work in pharma manufacturing. Before every new batch, GMP requires a line clearance: proof that nothing from the previous batch is left behind (tablets, capsules, labels, vials, cartons). Today it is done by eye, about an hour of operator time plus an hour of QA oversight per clearance, and a single missed tablet can become a product mix-up and a quality notification. Nothing on the line catches a clearance failure; people do. ClearLine puts fixed cameras on the line. Each camera remembers what clean looks like. When a scan runs, change detection flags anything new and locates it in the room; SAM 3.1 outlines what it is; Muse Spark judges it against that area's checklist (leftover / belongs / unclear, with a reason the operator can act on). Operators remove and re-check (the cameras confirm the spot is clean), spots the cameras cannot fully see are measured and listed as manual checks, and QA signs a timestamped clearance record. Built and scored in a simulated 16-camera packaging room, so every result is checked against where the simulator actually hid things. On the live Meta Model API, across 10 random clearances: 28 of 28 visible leftovers raised, 26 confirmed by the AI, 2 flagged unclear for a human, 0 dismissed, 0 AI errors, about $0.06 per clearance. Everything was built during the hackathon (Oct 26 - Nov 4). Solo, indie track, theme: Build for Your World.