Notices.
The side practice, on the side. The kinds of work the lab takes on for companies that want what it writes about done, not just described.
Independent evaluation
AI audit
Find out if the AI you already run deserves to stay.
An independent evaluation of the AI systems you already have in production. We measure reliability, hallucination rate, bias, security exposure, and real cost on your own tasks, not on a vendor demo. You get a written verdict for each system: keep, fix, replace, or remove.
Read the noticeTask mapping
Right-tool consulting
LLM, decision model, classic ML, rules, or a human. We map which.
Our flagship anti-hype engagement. We walk your workflows task by task and decide which intelligence belongs on each one: a general LLM, a decision model, classic machine learning, an honest rules-based system, or no automation at all. You stop paying LLM prices for jobs a lookup table does better.
Read the noticeOn-prem & private
Local AI implementation
AI that never lets your data leave the building.
Deployment of open-weight models on infrastructure you control: on-prem, in your private cloud, or fully air-gapped for regulated environments. Includes private retrieval over your own documents, so staff get AI answers grounded in company knowledge without a byte leaving your perimeter.
Read the noticeMachines & stack
Hardware + software
The machines, the GPUs, and the software to run them.
We spec, source, and configure local inference machines and edge devices, then install the complete software stack on top. GPU sizing is based on measured workload requirements, not brochure numbers, so you buy enough hardware and no more.
Read the noticeRouting layers
Hybrid architecture design
Every task routed to the cheapest intelligence that can do it well.
Design of a routing layer that sends each request to the right model: small local models for easy tasks, frontier APIs for the hard ones, decision models or rules where they beat both. You get the cost of small models with the reliability profile each task actually needs.
Read the noticePropose & verify
Human-in-the-loop workflows
AI proposes. Humans verify and decide.
Workflow design where models do the drafting, retrieval, and first pass, and accountable humans make the calls. We build the queues, review interfaces, escalation rules, and audit trails that make "a human checked this" true instead of aspirational.
Read the noticePolicy & practice
AI governance & training
Written rules for using AI, including when not to.
Usage policies, staff training, and explicit "when NOT to use AI" guidelines, written for how your organization actually works. We turn governance from a PDF nobody reads into habits: what may go into a model, what must be verified, and what stays human by default.
Read the noticeHow the lab takes work
- Measured, not claimed
- Reproducible tests on your tasks
- Independent
- No reselling, no vendor commissions
- Scoped in writing
- Acceptance criteria before spend
- Yours
- You own the deliverables