AI, compliant and production-ready

The entry format and the exit from uncertainty. One bounded examination of where AI is used, what it touches, and which of the EU AI Act and GDPR duties already apply.

You receive: a scorecard report — AI inventory with a risk class per use, shadow-AI findings, action list prioritised by exposure — written so it stands on its own, without me in the room.

public sector: fixed price for administrations, 15 working days

The training itself is the smaller half. The sellable half is the documented coverage: who needs what depth, who was trained, when, with which materials — for obligations in force since February 2025.

You receive: the evidence package — role-specific curriculum, participation records, a policy snippet naming who was trained in what. That artifact is what an audit asks for.

public sector: session rates for open cohorts on request

Strategy that ends in decisions, not decks — a phased roadmap, a target architecture (including the local/sovereign posture), governance with a named owner and a register. Bounded tool builds fit here as an option when a named artifact — a internal assistant, an eval rig, an analysis pipeline — is the deliverable.

You receive: per phase, one fixed artifact — the roadmap, the target architecture decision document, or the governance design. For administrations: the FRIA dossier, ready for supervisory review.

public sector: FRIA package (Art. 27): 2–4 weeks, fixed

For organisations past the first build: a standing technical authority who owns the eval bar, reviews vendor claims, and keeps governance alive as models and duties shift (Art. 26/27 from December 2027).

You receive: steady architecture and governance decisions on retainer — fixed monthly days, documented decisions, no licence resale, no vendor commissions.

public sector: procurement-safe retainer wording on request

Custom LLM environments (flagship case, write-up in preparation)

challenge

A client needed their own LLM environment — not an off-the-shelf chat wrapper: model selection, hosting posture, context engineering and integrations, end to end.

approach

Sovereign stack over open-weight models, knowledge-management layer on top, eval suite as the acceptance gate.

result

A running, owned LLM environment with measured quality at handover — documented so the client runs it without me.

Organisational knowledge systems (pattern case)

challenge

Knowledge scattered over silos, tools and heads; retrieval that cannot cite its sources is unusable in sensitive settings.

approach

Retrieval-first knowledge architecture — corpus curation, RAG with citations, knowledge-graph layer where relationships matter.

result

Answers with provenance and a system the organisation can extend itself.

Agentic tooling for BIM & AEC (pattern case)

challenge

Planning and construction data lives in document form (IFC, BIM execution plans) that LLMs cannot take in raw.

approach

Bounded parsers and validators — published openly — instead of one-off scripts: bim-parse, plan-diff, bim-execution-plan as reusable artifacts.

result

Open-source tools others now build on; the pattern generalises to any document-heavy domain.