Someone Has to Sign It
Accountability is the binding constraint once a retrieval-augmented system reaches production. The coda to the RAG series.
Turning rigorous research into intelligent systems.
Writing
Insights.
Essays on building and governing AI systems: as the techniques mature, the binding constraint shifts to judgement, governance and accountability. Written for the people who build, run and answer for them.
Start here
Three essays that together show governance judgement, regulatory awareness and the technical-to-current synthesis.
Foresight, Not Hindsight
Compliance is a floor, not a ceiling. What we owe the people whose lives sit inside the data, before harm arrives.
What I Recognised in 2012
Hinton, neural networks and the long road to deep learning: what the field looked like before the data and the GPUs arrived.
RAG in Practice
What quietly fails in retrieval-augmented generation, and how to catch it.
1. What Do You Want From RAG?
Getting clear on the job before reaching for the architecture.
2. RAG Does Not Stop Hallucination. It Relocates It.
Where the errors move to once retrieval is in the loop.
3. Cosine Similarity Is Flattering You.
Why retrieval scores look better than the retrieval is.
4. ROUGE and BLEU Will Not Save You.
The limits of convenient metrics for generated answers.
5. Do You Need a Vector Database at All?
When the infrastructure is the badge, not the requirement.
6. Your Knowledge Base Is an Attack Surface.
Retrieval opens a door; least privilege and provenance keep it watched.
7. Someone Has to Sign It.
The coda: who is accountable when the system is wrong.
GenAI & LLMs in the Enterprise
Why adopting AI well is more than calling an API.
It’s More Than Calling an API
The opener: you bought the capability in an afternoon and budgeted for none of the years it now has to keep working. Why adopting AI well is more than calling an API.
The Wrong Model for the Job
Using a language model for a task it was never built for, in a domain where being wrong costs money.
Agentic AI as the New Résumé Line
Agents touching production systems and customer data with no monitoring, audit logs or least-privilege thinking.
The MLOps Handoff
Monitoring is a modelling task, not a scripting task: drift, retraining triggers and data-quality decay can't be governed by someone who doesn't know how the model works.
The Shape of Optimisation
Where decades-old optimisation methods still run the world, and what is genuinely new.
Old Methods, New Combinations: Why 1980s Optimisation Still Runs the World
Simulated annealing, genetic algorithms and tabu search still run much of the infrastructure around us; the frontier is how we combine them.
Framework, Not Metaphor: What a Metaheuristic Actually Is
What a metaheuristic really is: a framework, not an algorithm, powered by memory and learning rather than randomness dressed in metaphor.
Neural Networks in Optimisation: Patience, Not Hype
Long before the 2012 breakthrough, optimisation researchers assessed neural networks honestly rather than dismissing or overselling. Why that patience aged well.
Metaheuristics at Fifty
EURO's fiftieth-anniversary review: fifty years on, the techniques are everywhere and the most interesting work is still ahead.
A Tension Worth Watching: Optimisation, Cryptography, and the Problems Between Them
The same intractable problems optimisation wants to crack are the ones cryptography needs to hold. A thirty-year view of that tension.
Foundations & Practice
The history, infrastructure and engineering judgement underneath today's tools, with the applied ML that puts them to work.
Why Financial NLP Needs Interpretability by Design, Not by Retrofit
With the EU AI Act in force, interpretability, uncertainty quantification and lifecycle monitoring are operational requirements, not optional extras.
When Data Outgrew the Machine: The Infrastructure That Made Scale Easy
Replicated storage, divided computation and rising abstraction turned big data from a specialist ordeal into something ordinary.
Reading the Signal in Time: How Forecasting Has Matured, and Why Finance Stays Hard
Forecasting now takes an afternoon, yet financial data, for reasons of codependency and regime change, stays among the hardest to forecast.
What Should You See Next? The Long Evolution of Recommender Systems
From which film you might like to the quiet logic ordering much of what we read, buy and believe.
The Long Way Round to a Simple Idea: Customer Value Modelling in Finance
A handful of undergraduate distributions predict customer value and defection with surprising robustness. Why simple models still hold their own.
Owning the Whole Problem: How Software Practice Changed
Agile corrected the real failures of older practice, but a particular kind of depth went with it, and it is worth recovering.
A Difference of Category: Hardware and Software Since the 1990s
Compute was a budget and memory was counted. Abundance lifted those constraints, and quietly removed an awareness scarcity once made compulsory.
Building Without the Scaffolding: What Finance Technology Achieved Before the Tools Caught Up
Without hardened libraries, databases or open source, 1990s investment banks built straight-through processing anyway.
From Caesar to the Internet: Robert Churchhouse and the Long Arc of Cryptography
Taught by Turing and Newman, eleven years on ciphers, a closing book from Caesar to the internet: the continuity of cryptographic security.
Requests for Comments: How Open Standards Built the Network We Trust
From plaintext Telnet to SSH and TLS: how open standards are made, and why their priorities shifted from connection to security.
Closer to the Machine: What Decades of Programming Languages Have Taught Us
Thirty years across Ada, C, C++, Java and Python: why every programmer should spend real time closer to the machine.
Earlier thought leadership
Externally published pieces, republished here with formatting updated for readability.
Diversity in AI
A thought-leadership article advocating inclusion and representation in AI.
Artificial Intelligence for Banks: The Revolution Starts Now
How AI will reshape knowledge work in banking, from research to compliance, much as robots reshaped manufacturing.
The Power of Predictive Analytics & Machine Learning
An ahead-of-the-curve case for machine-learning adoption in banking: mine your own transaction data.
Big Data: Going Beyond the Hype
An early commentary on data-driven transformation in banking: cutting through the hype to where data creates value.