The execution engine
Intelligence should compound, not reset.
Every project, every transition, every tool — the learning should stack, not scatter. The machine should handle the repetition. The person should handle the meaning.
A separate business · we are its first user
01What it is
An ontology-driven operating system, not a chat window.
VITAL Expert is a multi-agent platform built for regulated life-sciences work. It is not a general AI tool with a healthcare skin — the domain model is the product.
The ontology is the bridge that makes AI operational rather than conversational. Without a structure that encodes how the work is actually done, you can add as much model capability as you like and it will not produce a deliverable anyone signs.
Here you do not chat. You produce work products in a shared workspace, from templates that are reusable and shareable — so the second version of anything costs less than the first.
02Architecture
Three tiers, six questions, asked in order.
The sequence is not a diagram — it is how retrieval, orchestration, execution and measurement are actually wired.
Organisational knowledge, methods, institutional memory.
Context, governance, data classification and the audit requirements that apply. This tier is the difference between an assistant that is fluent and one that is grounded.
03Three ideas
Agents as workforce. Decisions as objects. Learning as a loop.
Agents sit under the same framework as people
They are not tools bolted to the side of an org chart. They carry competencies, are matched to jobs through the same intent resolution, and are managed, deployed and improved alongside human roles. One workforce, two kinds of member.
Every decision is a first-class object
Which agent was selected, which intervention level was chosen, what was known at the time, what alternatives were considered, and what resulted. That record is the foundation of both organisational learning and defensibility — and it is the thing that makes a second engagement cheaper than the first.
Outcomes feed back into every layer
Agent memory updates, priority scores recalibrate, the knowledge layer enriches. Organisations running this are not implementing AI — they are building a learning infrastructure where each cycle is informed by every cycle before it.
04The boundary
What it does not do, stated first.
It never produces a GxP work product, and it never writes content you file with a regulator. Not a configuration choice — a scope decision, and it holds regardless of what the platform is technically capable of drafting.
It never signs anything. The platform owns the run and the record. A named person at Crossroads owns the judgement and the signature, and those two things are deliberately held by different entities.
It is not represented as validated. Where a questionnaire asks, that is the answer. What we can describe is the control — what ran where, who touched it, and what the record shows. The full boundary →
05Position
Where it sits, honestly.
Against traditional consulting
Against AI-layered consultancies
Against AI-native platforms
Against technology builders
Against generic AI training
And what it is not. Not a chatbot layer. Not a dashboard product. Not a collection of pilots. It is the decision and execution layer for teams that need more than another AI surface.
06Pricing
Published there, not here.
VITAL Expert is a separate business with its own site, its own terms and its own price list — self-service through to enterprise. Consulting is scoped; products are priced, and mixing the two on one page is how a practice starts reading as a vendor.