Practical articles from the people who build the systems: what tends to go wrong, how we check it, and what we would do again.
Building a RAG chatbot that actually works in production means solving about 10 problems the tutorials skip — from hybrid retrieval and hallucination guardrails to conversation context and knowing when to skip the LLM entirely. Here's what I learned building one from scratch.
A retrieval-augmented chatbot is mostly a search problem with a language model on the end. Where these projects go wrong: chunking, retrieval quality, permissions, and having no way to tell whether the answers are any good.
Read articleForget the headline-grabbing demos. The changes that are reaching ordinary businesses this year are narrower, cheaper and less glamorous than the coverage suggests. Here is what looks durable, and what to be sceptical about.
Read articleUsing drone imagery and an object detector to flag crop that is ready to harvest sounds simple. The model is the small part. Labelling, image tiling, changing light and what counts as a mistake decide whether it is useful.
Read articleMost cloud bills are inflated by the same few habits: oversized servers, scaling on the wrong signal, and no one watching for odd spikes. Fix those first, then decide where machine learning actually earns its place.
Read articleMost dashboards get built, admired once, and then ignored. A visualisation strategy is mostly about deciding which questions deserve a screen, defining each number once, and making the result fast enough that people open it on Monday morning.
Read articleForecasting does not need a complicated model to be useful. A practical walk through setting a baseline, testing a forecast honestly, and telling decision-makers how uncertain it is.
Read articleDescribe what you are working on. We are happy to say what we would try first, even if it never becomes a project.