Separating fact extraction from the claims a profile eventually makes
Abstracts
Abstracts
A crawler agent prototype in this work fetches approved public links and pulls out timestamped facts, but it stops there. Turning a raw fact into a claim about someone's work is a separate step, handled deliberately rather than automatically. Keeping extraction and interpretation apart is a small architectural choice with a real payoff: it's much easier to trust a system's output when you can see where each piece of evidence came from before it becomes a sentence.