Joshua Woo
Founder, Perfraction
About
I'm a technology, privacy and AI lawyer who builds the tools I used to wish existed. Nine years in — most recently as Goldman Sachs' APAC lead for AI, cyber and privacy; before that, IP & TMT at Mayer Brown and Dentons, breach counsel on 15+ regional incidents, and lead technology counsel on a digital-bank build-out. I opened a terminal for the first time in February 2026, and I've been learning the way that sticks — getting my hands dirty, building. Some of it legal (my team's Margin won Most Innovative at the Singapore Academy of Law's 2026 Hackathon), some of it just for me: second-brain systems for daily life, and an interview tool that surfaces the gap between what someone says and what they actually do. One conviction throughout: the lawyer makes every judgment; the machine only extends how far that judgment reaches.
1 Project
Twinterview
A local-first app that talks with you to build a digital twin, then tells you how much of it to believe. Most "digital twin" tools hand you a form. Filling in boxes gets you the person you'd like to be on a good day: considered, tidy, edited. Twinterview interviews you by voice instead — it asks a question aloud, you answer aloud, and it follows up on what you actually said. Talking is looser than typing, and looser is where the real stories come out. It feels like being interviewed, not like completing a survey. It asks for specific episodes*rather than adjectives, separates what you say from what you actually did, mines your own writing and chats for behaviour you'd never think to report, and holds back a sealed set of questions to score how faithful the result is. Voice in, voice out, and every byte of your data stay on your machine. The only thing that leaves is the model call — and bring-your-own-key means you can point that at a local model too, so nothing has to leave your machine at all.