Zhen Yang Soh — AI Engineer

Zhen Yang Soh

As a kid, I was obsessed with space.

University of Bristol · 2017–2021

And so I studied rocket science.

That changed how I saw the world.

01First principles thinking

Then I saw AI for the first time.

It felt like magic.

“Any sufficiently advanced technology is indistinguishable from magic.”

Arthur C. Clarke

So I started to play with it.

Now · Kuala Lumpur

Today I build AI agents in insurance.

Three things I have built there.

01For insurance agents

An award-winning cross-sell platform that takes insurance agents from a customer's cover gaps to the products that close them.

  1. Concepts
  2. Prototype
  3. Agent workshops
  4. Iterate
  5. Live

I built a working prototype of the whole platform and tested it with agents before anything went to production. The roadmap followed what they asked for.

Awarded for this project

  • Best Use of AI for Customer Experience in Insurance

    The Digital Banker

  • Outstanding Digital CX Transformation in Insurance, Southeast Asia

    The Digital Banker

Find out more →

02For claims handlers

AI agents that assess an insurance claim, then pay the customer or hand it to a human for manual review.

One modular architecture, reused across claim types. Adding a new type of claim means writing a new guideline, not building a new engine.

Payouts that took weeks now take days.

Find out more →

1A policy agent reads the policy wording and pulls out the terms and payout rules.

reason act observe Claim files OCRevery page read Policy agentterms + calculation rules EvaluatorReAct loop Validatorchecks the result Payautomatically Claims handlerhuman in the loop Policy wording Guidelineper claim type swap in, not rebuild

2An evaluator works through the claim step by step, following a guideline written for its claim type.

4Handlers can open any decision and see why: every policy term, pass or fail, and the payout per item.

3Claims that need a person's judgement go to a claims handler, with the reasoning laid out.

03For contact centre staff

An assistant in Teams that answers product questions and shows the page it came from.

Staff get asked about every product, including legacy ones nobody remembers. It splits a question into its parts, rewrites each one several ways, and runs a hybrid keyword-and-vector search for every version. The results are merged and reranked, so the answer cites the exact page it came from.

Find out more →

All projects →