Operator

Priya Menon

Data readiness · London · 14 years

She tests the system on the files the vendor did not send.

FocusWhether the model survives messy, incomplete, real-world data.

• Messy data-room files
• Vendor vs real holdout
• Distribution shift
• Document AI systems

Closed work2026: Stopped a closeNorthlake diligence parser. Extraction held 0.91 on vendor data and 0.54 on the target’s actual files. The IC walked away.

Priya spent a decade building credit and document models that looked excellent on tidy sets and then failed the first week in production. She now does the inverse: she finds that failure before you buy or scale. PE diligence teams bring her in when the data room is a slide, not a system.