Retrieval over your own knowledge
Documents, tickets, policies and product data indexed and searchable with citations, so answers can be checked rather than trusted.
M3CA://AIOnlineRuntime: Retrieval + evalsGuardrails: On by defaultRegion: UK / EU
Demo
01Sector dossier
Make AI part of the company, not another tool on the side.
M3CA London · AI practice, designed and built from 35 Berkeley Square, Mayfair
M3CA AI is a standing team of retrieval engineers, model specialists, evaluation leads and trainers.
Relevant proof
What AI has actually shipped.
03 projects
- 01We integrate leading AI into the tools people already use, teach teams to work with it, and develop specialist or foundation models where the evidence supports the investment.
Cited
Answers tied to source material
Scoped
Access inherited from the asker
Tested
Evaluation before production rollout
The loop we build against
Six stages. Every one of them has an owner and an artefact before it ships.
Select a stage to read what we build there.
Documents, tickets and policies indexed with permissions intact
Permissions intact
Operating ledger
The paperwork is part of the product.
The controls that come with every deployment, written down before the first prompt is tuned.
- Provenance
- Every generated claim carries its source document and paragraph
- Permissions
- Retrieval runs as the user, never as a service account with wider access
- Evaluation
- Golden set from your real cases, regression-scored on every model change
- Cost control
- Per-team ceilings, model routing and cached retrieval
- Data handling
- No training on your data, UK or EU inference region, retention you set
- Fallback
- Deterministic path when the model is unavailable or below confidence
- Model route
- Integrate, fine-tune or train from scratch only when the requirement justifies the extra scale
What we deliver
Four outcomes, not four paragraphs.
Assistants with real context
Scoped to a team, a product or a patient pathway, with the permissions of the person asking, never a shared god-account.
Automation with a paper trail
Triage, classification, drafting and routing, each step logged so an auditor can reconstruct what happened and why.
Evaluation before rollout
Test sets built from your own cases, scored against a baseline, re-run on every model change.
A team that can use it
Role-specific training for leaders and staff, covering prompting, workflow design, verification, responsible use and human escalation.
Models matched to the mission
We integrate leading providers, fine-tune domain models or scope a foundation-model programme according to the requirement, data, risk and investment.
Products in this sector
All products
On M3CA TV
What we refuse
Three things we will not do.
- 01
Assistants without evaluation. If we cannot score it, we will not ship it.
- 02
Answers without provenance shown to a decision-maker.
- 03
Autonomous write access to a system of record on day one.
- 04
Training workshops without a working company use case behind them.
Disciplines
Switch sector
02HealthcareAdministrative software for healthcare providersOpen03EducationLearning that people finishOpen04Cyber & ComplianceSecurity you can evidenceOpen05MediaA working studio, not a stock libraryOpen06Managed ITThe layer everything else stands onOpen07Sous ChefFor the evenings that earn the candleOpen08OpportunityThe UK's media and IT opportunity engineOpen09MarketingLower customer acquisition cost, higher ROASOpen10TalentThe crew, verified and on the clockOpen
Start
Bring us the workflow, not the buzzword.
Tell us the decision people are slow to make and the documents behind it. We will tell you honestly whether a model helps.