When There's No Right Answer, We Ask a Model to Grade It Anyway

Most quality checks work by comparing an output to a known correct answer. In private markets, that correct answer often doesn't exist — a valuation is somebody's judgment call, and there's no market price to settle the argument. The common workaround is to have another model act as judge, deciding whether an AI-assisted valuation or decision looks reasonable. That approach holds up reasonably well when the judge has a ground truth to compare against; remove it, and judge reliability drops noticeably in exactly the conditions where organizations lean on it hardest. Yana Doroshenko will unpack why reference-free LLM-as-judge setups are weakest precisely where high-stakes, judgment-based decisions need them most, and what this means for teams building AI-assisted decision intelligence into private market workflows.
Format: TECHNICAL TALK
Location: ROOM C
Speakers: Yana Doroshenko
Level: Intermediate to Advanced
Audience: Registered participants