VendScan
Concept
VendScan turns the one photo a vending technician already takes (or should take) when opening a machine into a full physical-inventory scan. Snap a picture of the machine's interior during a restock or route stop, and VendScan's vision model reads slot-by-slot fill levels, OCRs expiration dates on visible packaging, and flags mechanical/sanitation anomalies (jammed coils, mold, pest activity, broken glass, coin-jam indicators). An autonomous background agent then updates that machine's live inventory profile, cross-references it with sales/DEX data, and auto-generates the next visit's restock manifest and optimal route order — surfacing only true exceptions (expired product, needs-repair, suspected theft/vandalism) for a human to review.
MVP
- Mobile app for technicians: open camera, snap one photo per machine (works with any smartphone, no new hardware).
- Vision pipeline (built on an off-the-shelf multimodal model) extracts: slot fill %, product identification, visible expiration dates, and anomaly flags.
- Backend agent auto-builds the next route's pick list per machine and re-sequences visits by urgency (near-empty or near-expiry machines first, full machines skipped).
- Simple ops dashboard for route managers showing auto-generated manifests, an exception queue (repairs, expired pulls, low-margin SKUs to cut), and a route history log so managers can audit exactly what the agent decided and why.
- Priced per-machine/month subscription for small-to-mid vending operators (50–500 machine fleets) who currently have zero visibility beyond coin/card sales totals.
Why AI, why now
Cheap, high-accuracy open multimodal models (e.g., GLM-5.3-class, MIT-licensed, low-latency inference chips) make per-photo slot-level analysis fast and inexpensive enough to run on every single route stop — a task that six months ago would have needed custom hardware or expensive proprietary vision APIs. This lets VendScan skip camera retrofits entirely and start from the technician's existing phone.
Unique selling points
- Zero new hardware — works with the phone technicians already carry.
- Blends vision sensing (what's actually in the machine right now) with an autonomous agent (decides the next best action) so operators get a "just tell me what to do" restock list, not another dashboard to babysit.
- Audit trail of agent decisions builds trust for operators skeptical of automation replacing route intuition.
- Clear expansion path: once adopted, upsell a cheap in-machine camera module for continuous monitoring (no technician visit needed to detect a stockout or failure), moving VendScan from an on-visit tool toward a true continuous watchdog product.
