Describe an outcome
Specific tasks create more meaningful neighborhoods than prompts such as “which AI is best?”
One task. Five relevant roles.
Use the source catalog as a discovery surface, not a leaderboard. Local MiniLM measures role-language proximity; you still verify current products separately.
Describe the work, not the brand
SOURCE CATALOG48 NAMED / 2 UNNAMED EXCLUDED
No current availability, link, price, privacy, benchmark, product quality, or time-savings claim is verified here.
AWAITING MODEL384D TARGET
STRONGEST ROLE MATCHNOT RUN
Run the local model to compute evidence.
Relevance starts evaluation.
Role language narrows the field. Cosine similarity exposes why. Pricing, privacy, reliability, and fit still require current verification.
Inspect the role
Every match preserves the exact short description supplied by the source instead of inventing feature claims.
Evaluate after discovery
A shortlist is useful because it reduces search space. It is not evidence of quality, safety, cost, or availability.