Case studies

Anonymous proof points, with the useful parts left in.

Public case studies are watered down on purpose: names, exact numbers, vendors, and proprietary topology are removed. What remains is the operating shape of the work and the decisions that made the system more buildable.

Industrial AI

Do you really want to build a foundation model?

An anonymized engagement with a team that wanted 'a foundation model' for large-scale sensor forecasting — and could not put a number on it. We priced the crazy, and the number surprised everyone in the opposite direction.

Turned an 'unpriceable' foundation-model ambition into a defensible CAPEX and OPEX number.
Produced a complete reference architecture: data plane, pre-training, per-tenant fine-tuning, evaluation gates, and serving.
Replaced a vague moonshot with a staged build-vs-adapt decision the board could actually vote on.

Read the case study

AI infrastructure

AI inference platform readiness

An anonymized planning engagement for a team assessing whether a large-scale AI inference platform could move from ambitious architecture to supportable production operations.

Reduced an oversized infrastructure concept into staged delivery decisions.
Separated capacity assumptions from release, rollback, and operating risks.
Created a board-safe narrative for cost, reliability, and governance tradeoffs.

Read the case study