What AI engineering teaches risk systems

AI engineering, the reason we have AI tools is fascinating. If we had a blank sheet of paper to build financial infrastructure and these insights it would be built fundamentally differently.

Observing the key engineering unlocks:

1. Much cheaper semiconductors

2. GPUs repurposed (optimised tensor libraries)

3. Stable-optimisations

4. Accessible data in massive amounts

5. Hyper scalers for training

6. Transform Architecture (the famous one)

7. Embedded vector representations

8. Human in loop alignment

9. Inference Engineering

What can we take from a Risk / Finance perspective.

Well we want estimator frameworks. If it isn’t important don’t calculate it.

We want vector embedding. Absolutely critical the key innovation for our space. We want data do training. We want human in loop architecture for training / testing.

So I’m not saying this is AI for Risk systems because that is a massive claim, but if we want self correcting deterministic outcome risk systems in real-time we need to drop the entirely Markov chain distributions systems outside of testing.

We need estimator frameworks. We need to rethink clearing.

Why, well if tokenised assets are going to be used as collateral it adds no value if they are always locked waiting for risk systems to batch before they can move then we have gained almost nothing.

We need this. We need this as a horizontal service to support all the innovations on the matching layer and the price discovery.

Stablecoin / tokenised collateral clearing built on AI estimator formats. That’s where we are with DCN.

This is what we do.

This first appeared on LinkedIn on 1 April 2026. If you want to comment or discuss, that’s the place.