Category: AI & Data
-
Vector databases: when your AI feature actually needs one
A vector database is not the default answer for every AI feature. Here is how to work out whether semantic…
-
RAG, fine-tuning or prompt engineering: choosing how to customise an LLM
A practical framework for choosing between prompt engineering, retrieval-augmented generation and fine-tuning when commissioning an AI feature.
-
Data quality: the questions to ask before you commission an analytics project
What to check before commissioning an analytics project: metric definitions, validation rules, and how you’ll know later if the numbers…
-
Explotar tus datos sin tener un equipo de datos
Si dos personas de tu empresa dan dos números distintos para la misma pregunta, el problema no es la herramienta.…
-
Testing AI features before launch: how do you know they actually work?
AI features fail quietly, not with error codes. Here is how to build an evaluation process that catches that before…
-
La IA en tu empresa empieza por un problema aburrido, no por un chatbot
La IA no empieza por un chatbot. Tres sitios donde paga de verdad en una pyme, los cuatro puntos donde…
-
Data warehouse or data lake: choosing a data architecture for a growing business
A practical framework for choosing between a data warehouse, a data lake and a lakehouse, based on data shape, query…
-
Controlling LLM API costs in production: the levers that actually work
Token spend climbs fast once an AI feature reaches real usage. A practical look at caching, routing, batching and budgets…
-
Transaction categorisation in personal finance apps: build, buy, or blend
Why categorising bank transactions is harder than it looks, and how to weigh rules, in-house machine learning, and a third-party…
-
On-device or cloud: choosing where your AI inference runs
A practical framework for deciding where AI inference should run: on-device or cloud, weighed against latency, cost, data handling and…