Blog

AI rarely fails because the model is “dumb”, it fails because the knowledge it’s pulling from is messy. If you’ve ever had an AI answer at work that sounded confident but made you think “hang on… is that actually true?”, you’ve hit the real issue: shaky inputs create shaky outputs. In this short video I explain why, in 2025/2026, enterprise AI is only as trustworthy as the knowledge base behind it, and why duplicates, contradictions and outdated docs quietly turn into errors,

Here’s the reality in 2025/2026: there isn’t one universal “where does AI gets its facts from?” because most frontier labs don’t fully disclose their training mixes anymore. But we can  anchor this in what’s publicly documented and what regulators/researchers keep pointing at. You know what people get wrong about AI? They think it “looks things up” like a clever librarian. Most of the time it doesn’t. Most of the time it’s answering from whatever got baked into it during trai

If you’ve been following the AI boom, you’ve probably heard the term Vector Database  tossed around. It sounds like something straight out of a physics textbook, but it’s actually the "secret sauce" making modern AI - like ChatGPT or advanced recommendation systems - work so efficiently. But what are they, and why should you care? Let’s break it down without the jargon. What is a Vector Database, anyway? Traditional databases (the ones we’ve used for decades) store informatio

You ask an AI a question (ChatGPT or other - they're all as guilty as one and other) and it replies instantly with something that sounds like it came from a well read expert who also happens to be polite, structured, and suspiciously confident. It's fluent. It's tidy. It even gives you bullet points. And then you discover one awkward detail. It is wrong.

We’ve all spent the last 18 months talking about “AI for customer service” like it’s a magic trick. Better bots. Smarter assistants. RAG for everything. And yet… handle times barely move, recontacts creep back up, agents still ask in Slack:  “Which article is actually correct?” It’s not that the AI is bad. It’s that we’re asking it to reason on top of knowledge that’s fundamentally broken. Messy, contradictory, duplicated, outdated, scattered across Confluence, SharePoint, Se

"Ontology is lining up to be the buzzword of 2026" I'm hearing that everywhere now, but really? Palantir’s rise has put ontological modelling back in the spotlight – their Foundry platform is built on it. Microsoft is now moving ontology into Fabric. The race is on. It makes sense. Ontologies give generative AI something it desperately needs:  grounding . LLMs are brilliant at pattern-matching and language, but terrible at enforcing logic. They’ll happily smooth over contradi