How it works. An agent queries public sources on a scheduler, discards most postings with deterministic filters before calling the LLM, analyses the rest and fills in forms in browser sessions a person can supervise.
Result. About 97% of postings discarded by deterministic filters before the LLM is called.