AI

AI tools can produce data analysis that looks authoritative without being verifiable. Whether you're the analyst or the decision-maker relying on the output, this post outlines some practical principles for making AI-assisted analysis trustworthy.

Understand the generation step of RAG: how LLMs use augmented context to produce grounded responses, how to enforce structured outputs with Pydantic, and how to evaluate response quality.

Understand the augmentation step of RAG: how retrieved documents are structured into prompts, how metadata and citations improve response quality, and how to avoid common pitfalls like prompt injection.

Understand the retrieval step of RAG: Learn how database queries, keyword search, vector search, and hybrid approaches find the right information to ground your LLM responses.

Traditional cloud SLAs measure API uptime, not behavioural stability. This post talks about how silent AI model drift can quietly break your agents and the questions you might want to ask vendors.

AI is like introducing a powerful new instrument to an orchestra. It creates possibilities that didn't exist before. But it still requires musicians who can read music, a conductor with a vision, rehearsal time, and the discipline to play together. An orchestra that lacks these fundamentals won't be saved by a new instrument — they'll just make new kinds of noise.

Top-down AI strategy and bottom-up experimentation both fail alone: leading organisations combine them to drive real business results.

A simple mental model for making sense of the AI-assisted coding landscape: four layers, four decisions.

Shift from prompt engineering to AI programming by applying rigorous software engineering principles to your LLM integrations.

What is RAG? Learn how RAG combines retrieval, augmentation & generation to ground GenAI responses in your data while reducing hallucinations & improving accuracy.

We have a wide range of computational mechanisms at our disposal, some of which emerged thanks to recent advances in AI. In this post, we look at the kinds of workloads that can take advantage of these.

Thanks in part to recent advances in AI, we have a range of computational mechanisms at our disposal. However, certain universal truths apply to all of them.

AI's current hype cycle has driven massive hardware investment. Even if AI disappoints, those hardware capabilities remain. What else can we do with them?

Explore AI-powered self-service reporting with tools like Copilot in Power BI and Microsoft Fabric, balancing benefits and pitfalls.