LLMs know the world. They don’t know your world. The semantic layer — ontology, router, retrieval stack, context assembler, and grounding layer — is the missing infrastructure that makes agents trustworthy in specific domains.
The U.S. healthcare system loses up to $300B a year to fraud. Static rules engines can’t keep up. Agentic AI — with autonomous orchestration, structured knowledge layers, and closed-loop integration — is changing the game.
AI isn’t a productivity tool — it’s a civilisational shift on par with electricity and the internet. Here’s what the new world looks like and how to position yourself in it.
My understanding how temperature, top-k, and top-p decoding work in large language models like GPT.
For the last two years, I have taken a New Year resolution every year and worked hard to stick to it. Both years turned out to be an astounding success. So, for 2025, I decided to embrace a new resolution: to write more! Over the past few years, I have accumulated a wealth of experience but have had little time to document it. In the last three years, I have successfully shipped eight different products, spent countless hours diving into options theory math, skilled up in Generative AI, experimented extensively with LLM applications, built websites, streamlined applications, and much more. The learnings have been immense, but sadly, much of it remains undocumented. ...