Product inventory auto-tagger
Automatically classify and tag your product catalog using lightweight NLP inference running entirely on the edge. No external API calls, no per-request billing.
What this recipe solves
Manual product tagging is slow, inconsistent, and doesn't scale. This recipe deploys a compact transformer model to Vercel Edge Functions, processing product titles and descriptions in real time. Tags are deterministic, cacheable, and computed at the edge — no round-trip to a GPU cluster.
Input
Array of product objects with title and description fields. Up to 100 items per batch.
Output
Each product enriched with a tags array and category string, plus confidence scores.
Model
DistilBERT quantized to INT8 (~87 MB). Loaded once at edge function cold start, reused across invocations.
Latency
P50: 28ms, P99: 62ms for a 50-item batch. Measured on Vercel Edge in fra1 region.
Ready to ship this recipe?
Browse more recipes →