GearWitness: Privacy-First Local AI Return Inspection for Equipment Rentals

Description and use

GearWitness reduces return inspection time, lowers disputes, and ensures all accessories are accounted for while keeping all data on the shop’s machine.

Problem

Independent camera‑rental shops and similar operators struggle with long, error‑prone return inspections that lead to customer queues, disputed charges, and missed accessories.

Outcome

GearWitness reduces return inspection time, lowers dispute rates, and ensures all accessories are accounted for while keeping all data on the shop’s machine.

Workflow

• 1. Capture checkout and return photos
• 2. Align images with OpenCV ORB keypoints, RANSAC homography, and filtering
• 3. Crop aligned regions and send to local Gemma 3 via Ollama
• 4. Receive structured proposals (condition, severity, confidence, description, rationale)
• 5. Operator reviews each proposal: Confirm Damage, Dismiss Wear/Glare, or Mark Uncertain
• 6. Reject final approval until all proposals are resolved
• 7. Export manifest JSON, printable HTML, or vector PDF report with SHA‑256 integrity digest

Evidence

GearWitness aligns checkout and return photos, proposes evidence-backed changes, reconciles the kit inventory, and requires a human decision before an inspection can be finalized.

Source

The public Render deployment deliberately identifies itself as Deterministic Fixture mode. It demonstrates the complete operator workflow without pretending that the free hosted container has a local Ollama runtime.

Source

OpenCV ORB keypoints, Hamming-distance matching, RANSAC homography, perspective warping, bilateral filtering, luminance subtraction, and Otsu thresholding.

Source

Aligned crop pairs are sent to Gemma 3 4B through Ollama in local mode.

Source

Every proposal has exactly three operator actions: Confirm Damage, Dismiss Wear / Glare, Mark Uncertain.

Source