AI & Applied Intelligence

AI features that still work six months later.

Answering questions from your own documents, assistants that carry out tasks, voice and chat interfaces, and sorting or pulling information out of messy data. We run our own AI products, so we will be straight with you about what AI will and will not do.

ClaudeOpenAIRAG & vector searchAgent orchestrationEvaluation harnessesSpeech pipelines

What you get.

01Use-case assessment with an honest build / buy / do-not recommendation

02Retrieval architecture over your proprietary data

03Prompt, tool and agent design with versioning

04Evaluation harness with regression testing on real cases

05Guardrails, fallbacks and human-in-the-loop escalation

06Cost, latency and quality monitoring in production

You have been told to do something with AI, but nobody has said what

A prototype impressed everyone and then failed in production

You hold proprietary data that a generic model cannot reason about

Not sure this is the right capability for your problem? A conversation usually settles it in half an hour.

Book a free call

How this work usually runs.

Qualify

Roughly half of proposed AI features are better solved conventionally. We say so before you spend the budget.

Ground

Retrieval and tool access over your own data, because a model without your context is a very confident stranger.

Evaluate

An evaluation set built from real cases, run on every change. Without it you are shipping vibes.

Operate

Cost, latency and quality monitored continuously, with fallbacks for when the model is wrong, which it will be.

Transport and marketplacesMuvmentThree different groups, each needing their own product, all depending on the same cars and the same calendar.

Questions we get asked.

Regularly. A deterministic rule that is right every time beats a model that is right most of the time.

An evaluation set built from your real cases, run on every change, with the numbers shared. Not a demo.

NextCloud, DevOps & Infrastructure

See all ten