Ideas
10posts
Re-architecting Next Trains
One of the things I have learned from working in data and analytics is that continuous improvement never really ends. Even when something works, there is always another question worth asking: Can I do this better?
Beyond Guardrails: Why Enterprise AI Needs a Semantic Control Plane
Enterprise AI needs more than guardrails. This article explores ontology, governance, and a semantic control plane for safer, context-aware agents.
What Ninja Tag taught me about technology strategy
How a kids’ obstacle game revealed a common failure in enterprise technology decision-making.
Next Trains
You just want to know when the next train is. It should take three seconds. Instead it takes three apps, two page loads, and a mild existential crisis on a rain-soaked platform.
Return of the data centre
AI is driving a quiet shift back to hybrid architecture, with critical workloads and sensitive data moving on‑premise while the cloud remains the lab for experimentation, scale and convenience.
What if hiring worked like a chess clock?
The Pitch Switch is a fast, reciprocal hiring night: five‑minute candidate and employer pitches, visible countdowns, decision‑makers in the room, and “fast‑track” tokens to convert interest into scheduled interviews on the spot.
Whose Fault Is It, Really? A journey through fragmentation, blame, and performance
From UK railways to SaaS, fragmented responsibility leaves customers proving blame while providers cite labs and SLAs. This essay argues for shared observability, transparent hand‑offs, and performance targets users actually feel.
Can AI Tell What’s Funny? Teaching Humour to Machines
Can AI understand humour? A practical look at wordplay, absurdity, and how language models generate jokes—plus why human verification matters for writing and code. Clear takeaways for using generative AI to communicate better without losing judgement.
Don’t Train On Me—A Practical Way to Attach Permissions to Your Digital Content
Attach clear, portable permissions to your content—no LLM training, share rules, and version history—using standard web metadata plus a tiny blockchain anchor. Same solution, two explainers: a human‑friendly overview and a technical deep dive. Provenance without the crypto headache.