A developer launched a free, no-login web app that helps NYC shoppers stack card cashback, weekly coupons, and CPG rebates across roughly 690 stores. Users enter items — comma-separated for multiple products — or hand the list to an AI shopping assistant powered by a fine-tuned Llama model. The tool surfaces combined discounts that most consumers miss because they stick to a single store and rarely hunt through fragmented deal sources.

The builder acknowledges two hard constraints: coverage gaps and data freshness. Pricing feeds come from multiple retail APIs and scraped circulars, none of which update on the same cadence. Some stores push daily price files; others refresh weekly or only when a circular drops. The developer is asking the community which failure mode hurts more — stale prices on a broad catalog or fresh prices on a narrow one.

Practical takeaway: anyone building on messy multi-source retail data should instrument freshness timestamps per source and expose them to users. A "last updated" badge per store lets shoppers decide whether to trust a $3.49 chicken breast quote from three days ago. The Llama integration also shows how small models can replace rule-based parsers for normalizing messy product names across circulars.