AI architecture · For IT leaders
The AI harness for quality and regulatory teams. Agents orchestrated for every business task.
A fleet of autonomous AI agents, orchestrated for each request according to the context of your company, your products and your standards. Precise, repeatable and sourced results.
Live
Every request organises its own agent swarm.
The harness loads the context and plugs in the right connectors. The orchestrator then creates the agents it needs, which in turn create their own, exchange and check each other all the way to a sourced result, across corpora of thousands of documents.
Example requests
Live orchestrationExample 1/4Assess the technical file of product B before submission.
EUDAMED
Score per requirement · 23 sourced findings · action plan ready
Key concepts
Harness, agents, orchestration: what they mean at MAC.
The vocabulary of agentic AI, applied to medical device quality and regulatory affairs.
- AI harnessagent harness
Everything around a language model that turns it into a reliable tool: knowledge, tools, workflows, guardrails, memory and evaluation rules. The model reasons; the harness decides what it sees and what is expected of it.
At MACThe harness carries the regulatory know-how: MDR, IVDR, ISO 13485 and applicable standards broken down requirement by requirement, a map of your documentation, your target quality level.
- AI agentautonomous agent
A program that relies on a language model to complete a task end to end: it reads, uses tools, checks and returns a result, without step-by-step guidance.
At MACEach agent has a narrow, well-defined role: assess one requirement, check the consistency of two documents, draft a finding, compare an EUDAMED declaration with your file.
- Agent orchestrationmulti-agent system
Coordinating several agents to complete a complex task: splitting the work, assigning it, running in parallel, consolidating the results.
At MACFor each complex request, MAC assembles the right team of agents for the user's need and context, runs them in parallel, then consolidates and checks the results before returning them.
- Agent fleet or swarmagent swarm
A set of specialised agents mobilised together, whose composition changes with the task at hand.
At MACThe fleet covers the quality system, technical files, regulatory watch, audit nonconformities, device changes and post-market follow-up.
Why a harness
The model is an engine. The value lies in the harness.
My Audit Corner is not a chatbot sitting on top of an LLM. Large language models are improving fast and look more and more alike; on a regulatory task, what separates a plausible answer from a correct one is everything around the model. That is where MAC's know-how lies, built by medical device practitioners.
A general-purpose LLM on its own
- A long, mixed context you have to build yourself
- Answers that vary from one run to the next
- No knowledge of your standards or your target quality level
- Sources to check by hand
- All the integration work still to do
The MAC harness
- Standards broken down requirement by requirement
- Granular agent roles, designed by domain experts
- A short, focused context calibrated for every call
- Every finding linked to its source document and passage
- Business value straight away, nothing to build on top
Orchestration
One request, a swarm of agents, one consolidated answer.
- 01
Understand the request
MAC identifies the business task and its context: device, class, applicable standards, documents involved, target quality level.
- 02
Assemble the swarm
The orchestrator picks the agents needed and splits the work into unit tasks.
- 03
Run in parallel
Each agent works on its own scope, with its calibrated context and its tools.
- 04
Consolidate and check
Results are cross-checked, inconsistencies flagged, every finding linked to its source.
- 05
Deliver
Scores, findings, action plan or ready-to-use drafting, which your team reviews and approves.
Repeatability and precision
Same documents, same requirements: same result.
It is the first question asked about generative AI, and a prerequisite in quality. MAC's repeatability does not come from tuning the model: it comes from the design of the harness.
Fine-grained agents
Each agent performs a narrow, well-defined action. It does not get lost in an oversized context and does not try to do everything at once.
The same context on every call
For a given task, the harness gives the model the same short, focused context. Same context in, same answer out.
Sourced answers
Every finding points to the document and passage that support it. Nothing is stated without verifiable evidence.
Humans stay in control
MAC assesses, proposes and drafts. Your team validates, arbitrates and decides. AI assists; it does not sign off for you.
Two analyses of the same large document corpus, against the same requirements and with unchanged documents, produce the same scores.
Confidentiality: your documents stay yours.
Hosted in the European Union on Google Cloud Platform, data encrypted in transit and at rest, isolated environment. No Training Data policy: no model is ever trained on your documents. ISO 27001 certification journey under way, penetration test attestations available to your CISO. GDPR compliant, EU AI Act requirements anticipated.
IT integration
MAC fits into your environment, with no migration.
MAC connects to the spaces where your quality and regulatory documents already live. No data to migrate: MAC reads, maps and analyses your documentation where it is. It is up and running within hours. An architecture question specific to your IT? Our technical team works it through with yours.
SharePointGoogle DriveConfluenceJiraYour questions, our answers.
What is an AI harness?
An AI harness (or agent harness) is everything around a language model that turns it into a reliable tool: knowledge, tools, workflows, guardrails and evaluation rules. At My Audit Corner, the harness carries medtech regulatory know-how: MDR, IVDR, ISO 13485 and applicable standards broken down requirement by requirement, a map of the customer's documentation and the target quality level.
Is My Audit Corner a multi-agent system?
Yes. MAC orchestrates a fleet of autonomous, specialised AI agents. For each complex request, it assembles the agent swarm suited to the task and the user's context, runs the agents in parallel, then consolidates and checks their results before returning them.
Are MAC's results repeatable?
Yes. Each agent performs a narrow action, with a short context calibrated by the harness: for a given task, the model always receives the same context. In production, two analyses of the same large corpus, against the same requirements and with unchanged documents, produce the same scores.
Why not build this ourselves with an LLM or an agent platform?
It works for a demo; it is much harder in production. MAC's value lies in the harness: how standards are broken down, how tasks are split and what each agent does, all designed by medical device practitioners, plus the ability to process thousands of documents. Your teams get the business value directly, with no development project to run and no standards library to keep up to date.
Are our documents used to train models?
No. No Training Data policy: your documents, patents and technical files are never used to train a model, for you or for anyone else.
Where is the data hosted?
Exclusively in the European Union, on Google Cloud Platform, with data encrypted in transit and at rest. MAC is on an ISO 27001 certification journey and makes its penetration test attestations available to your CISO.
Is MAC compatible with the EU AI Act?
MAC is designed in anticipation of the EU AI Act requirements: results traceable to their source, human oversight of decisions, transparency about the role of AI. MAC is also GDPR compliant.
How does MAC integrate with our IT systems?
MAC connects to your existing document spaces (SharePoint, Google Drive, OneDrive, Confluence, Jira, Odoo, TraceX) with no data migration, and it is up and running within hours. For any architecture question specific to your IT, our technical team works directly with yours.
Questions about the architecture? Let's talk.
Our technical team answers your requirement grids, shares the security documentation and shows you the MAC agent fleet at work.