Consulting
Applied AI, by the people who run the cluster
Orchestration, RAG, local inference and evaluation. We build the same systems we operate, which is why the advice tends to be about throughput and failure modes rather than about prompt wording.
What we do
Four kinds of engagement
Most projects are one of these, or one leading into another.
Local inference
Specify, install and tune a cluster in your own environment — model selection, serving stack, quantisation, throughput.
- Hardware sizing
- Serving stack
- Quantisation
- Benchmarks
RAG and retrieval
Get answers grounded in your own documents, with retrieval you can inspect and evaluation that tells you when it regresses.
- Chunking and indexing
- Hybrid retrieval
- Grounding
- Eval harness
Agent orchestration
Multi-step agents that call your systems: tool design, guardrails, retries, and the boundary between deterministic code and the model.
- Tool design
- MCP servers
- Guardrails
- Cost control
Evaluation and monitoring
Know whether a change helped. Task-specific evals, regression suites, and monitoring that surfaces drift before a user does.
- Golden sets
- Regression suites
- Drift alerts
- A/B harness
When to call
The four situations we see most
A demo that will not survive production
It answers well on the five documents you tested and falls apart on the five thousand you have.
No way to tell whether a change helped
Prompts get edited, quality moves, and nobody can say by how much or in which direction.
An agent that fails in surprising ways
Loops, runaway spend, or a tool call that does something nobody intended at 3am.
Data that cannot leave the building
The obvious provider is off the table for reasons that are legal, not technical.
Process
Three steps, no retainer to start
- 01
Scope
A call to establish what you are running, what it costs today, and whether we are the right answer. Often we say no.
- 02
Benchmark
We run your workload — your prompts, your documents — on the cluster and show you throughput, latency and quality against your current setup.
- 03
Deploy
Reserved capacity on our cluster, or the same stack installed in your racks. Either way you leave with something running.
Tell us what is not working
A scoping call costs nothing and quite often ends with us saying the problem is smaller than you thought.