Compliance AI, Improv, and Running Your Own Models
Andy Green — founder of Three Points Compliance in Hong Kong, building AI tools for SFC-licensed firms to monitor employee trading and conflicts of interest. PhD. Thirty years across GSM anti-fraud systems at Neural Technologies, the early terminal build-out at the Hong Kong Exchange, and a string of startups. Works with his business partner Trini So, a long-time compliance specialist. Long-time AI tinkerer, improv and comedy fan.
Chapters
- 0:00Intro
- 0:14Are we buddies now?
- 1:49Who is Andy Green?
- 3:40From electronics to machine learning
- 4:36One of the UK's first ML master's courses
- 5:32Building data distribution for the stock exchange
- 8:25Designing the wireless betting network
- 10:22Before GPRS — DHL, UPS, TNT, FedEx on the network
- 15:05Emergent behaviour, and why it surprises us
- 18:05Biologically motivated computing
- 20:54Neural networks as fringe research
- 24:38Three Points Compliance — monitoring employee trading
- 26:44Orchestration and a multiplicity of experts
- 30:48GSM anti-fraud at Neural Technologies
- 33:47The Cyberport incubator
- 36:39Working with Trini So
- 37:34The 80-20 rule in compliance
- 42:17Proving you followed the rules
- 45:08Improv, and why comedy helps
- 48:46Learning to program on paper tape
- 51:35The day the PDP-11 arrived
- 55:30How big does a model need to be?
- 57:24Why smaller models matter
- 58:19The information never leaves the office
- 1:02:12Closed models and access from Hong Kong
- 1:03:10What a neural network actually does
- 1:06:56Six months on — 'I can just wing it'
- 1:07:41How much RAM do you actually need?
- 1:10:41A new industrial revolution
- 1:11:38Why Andy is optimistic
About This Episode
Andy Green has spent thirty years building systems that cannot be allowed to fail — GSM anti-fraud for mobile networks, the early terminals at the Hong Kong Exchange, and now Three Points Compliance, which builds AI tools for SFC-licensed firms. He argues compliance is the one place where a plausible answer is worse than no answer, and explains how you build AI that knows when to stop. Also: why he runs Qwen at home on modified hardware, why 'never' is a dangerous word, and what improv comedy taught him about engineering.