Customer story · Medical devices

A medical device manufacturer drafts its regulatory file with eight agents. A person reviews and signs every page.

Every new device this company brings to market needs a full FDA design-control file and a 510(k) submission. Its small R&D team wrote all of it by hand. Now a private Brain, running in the company's own Microsoft cloud, drafts each stage from the company's own templates and procedures.

Client names available upon request.

8
specialist agents, one per stage of the design-control file
Research to 510(k)
100%
of drafts reviewed, edited, and signed by a person
Under existing document control
130+
approved company documents studied before the first draft
Templates, SOPs, past files
Own
cloud, storage, and documents, separate from any model
Swap models, keep the data
01The challenge

A small R&D team, and a full regulatory file for every device.

Bringing a device to market means a project plan, an initial risk analysis, design input requirements, validation protocols and reports for cleaning, packaging, performance, and sterilization, and finally the 510(k) summary. Each document follows the company's own procedures, templates, and numbering rules. Each one was written by hand, by a team that would rather be in the lab.

The work is repetitive in structure and unforgiving in detail. Getting a reference wrong, or missing a requirement that should trace through to a test, costs weeks.

02What we built

Eight agents, one for each stage of the file.

Research agent

Pulls public FDA data for the device: adverse-event reports, prior clearances and predicates, classifications and product codes.

In build

Project plan and design inputs

Drafts the project plan and the design input requirements from the device documentation and the applicable standards.

In build

Risk analysis

Drafts the failure modes and effects analysis, one line item per process step, in the company's template.

In build

Protocols and reports

Drafts validation protocols, then the reports, in the company's exact formats and numbering.

In build

Traceability and 510(k)

Traces every requirement through to its test, then assembles the submission package for review.

In build

Every draft lands in a review queue. The R&D team edits, approves in sequence, and signs, exactly as before.

03Privacy and compliance

Patient data stays out. A person signs everything.

The regulatory file is built from device documentation, public FDA data, and the company's own standards, not patient records. The platform is designed to keep protected health information (PHI) out of scope.

The constraint
What most companies do
What we built
Protected health information
Paste documents into a public chatbot and hope nothing sensitive is in them.
No patient records in scope. If a future workflow touches PHI, such as complaint files, it runs under a signed Business Associate Agreement (BAA).
Where the data lives
A vendor's shared cloud.
The company's own Microsoft cloud and storage. Changing the AI model does not move or touch the data.
Safeguards
Whatever the tool ships with.
Encryption in transit and at rest, role-based access, audit logging, and the minimum-necessary rule.
Who approves
Let the AI finish the document.
Every output is reviewed, approved in sequence, and signed by a person under the company's document control.

HIPAA is a set of rules a company and its vendors follow, not a certificate. We build to it and sign a BAA wherever PHI is involved.

04The result

The company owns the platform, the data, and every document.

8
agents drafting the design-control file, stage by stage

Each draft follows the company's own templates and procedures and is traceable to the sources it used. The team reviews and signs instead of starting from a blank page.

What changed

From blank page to review

The R&D team starts each document from a draft built on its own approved files, so its time goes to judgment and testing rather than formatting and cross-references.

What's next

The company's own team takes the wheel

Because the code, cloud, and data are the company's, its own team can keep developing the platform. That is the point of owning it.

Keep reading

More customer stories.

Customer, staff, product, and program details have been removed to protect the client's confidentiality. Client names available upon request.

Your story next

Find out what a private AI Brain would give back to your team.

In one strategy session we map where a Brain would return the most time, what we'd build first, and how we'd measure it from month one.

Book a strategy session

Prefer email? hello@nbrain.ai