1. Structured catalog
We maintain explicit fields for tasks, categories, platforms, features, pricing labels, summaries, and last-verified dates. The catalog is the factual boundary for the finder.
2. Deterministic filtering
The user’s query is matched against task, category, feature, platform, and phrase signals. This produces a shortlist of roughly 10–15 candidates before a language model sees the request.
3. LLM ranking and explanation
The model is instructed to rank only the supplied candidates. It cannot add products to the shortlist. Its job is to reason about fit and communicate trade-offs.
4. Commercial separation
Sponsored, affiliate, or publisher-promoted placements are labeled and displayed separately from organic recommendations. Commercial relationships do not change deterministic scores.
5. Verification cadence
AI products change quickly. Each profile includes a catalog update date and links to the official vendor. Users should confirm current pricing, availability, privacy, and limits before purchase or deployment.