The problem What it does How we work FAQ Let's talk →
ES EN
IVD · Medtech · Regulated environments

AI that reduces the documentation burden in regulated environments

We help in-vitro diagnostics and medtech companies take the documentation work that adds no product value — review, cross-checking, traceability — off their engineers' hands, using AI deployed on your own infrastructure. No data ever leaves your network. Your quality system stays untouched.

Review cycles burn engineering hours that should go into the product

In regulated R&D teams, a substantial share of engineering time goes into documentation: requirements and their traceability, verification protocols and reports, design documentation, technical file contributions. The biggest cost is not the writing: it is the review cycles. Documents that go back and forth two or three times between issue and sign-off because of mechanically detectable defects — inconsistencies between protocol and report, requirements without coverage, unrecorded changes. Every cycle burns author and reviewer hours, and adds weeks to project milestones.

General-purpose AI tools do not solve this in your sector, for two reasons: they know nothing about the formats and logic of a regulated quality system, and they require sending confidential development documentation to third-party servers, running on models that change without notice and therefore cannot be validated.

The system assists the engineer before the document enters the formal review loop

What it does. The system reads the agreed documentation scope — read-only — and generates findings reports for the authoring engineer, before the document enters the formal review loop:

What it does
  • Protocol–report cross-checking: consistency between what was specified and what was reported.
  • Traceability verification: requirement coverage by verification cases, dangling references.
  • Version comparison: undocumented changes between revisions.
  • Completeness review against your QMS checklists.
  • Transformation of existing content: test results into report format, traceability matrices, master protocol adaptation.
Every finding comes with its location and literal quotes, and checking it takes a person seconds, not hours. That is the design criterion: we only automate where verifying costs a fraction of what finding by hand would.
What it doesn't do, by design
  • It does not create, modify or approve any record in your quality system: it connects read-only.
  • It does not draft critical content. The author and approver of every document remain exactly who they are today.
  • It does not replace any step of your review loop, and it does not shift responsibilities.
  • It does not catch 100% of defects. No tool does — automated or human — and you should distrust anyone who promises otherwise. That is why it sits before your review, not instead of it, and why it is validated with measured metrics, not promised ones.

Your documentation does not need to leave your company

Your development documentation does not need to leave your company. We work with open-source AI models deployed on a server you own, inside your network: when the system analyses a document, the document travels from your repository to your server and back. No external calls, no telemetry, and the option to operate with no internet access at all. Your IT team can verify it by inspecting network traffic: the guarantee is architectural, not contractual.

You define the access perimeter: the system only indexes the folders or projects agreed, and every query is logged. Usage metrics are aggregated by process — never by person.

A tool your quality department can assess

The tool comes with its validation file prepared, so your quality department can assess it under your own software validation procedure: documented intended use, frozen model version (no silent update can ever alter what you validated), a test bench of documents with known planted defects and a measurable detection criterion, and change control with a regression suite for any update. We include a reasoned classification of the tool under the EU AI Act.

Three stages of growing commitment

01
3–4 weeks · €5,900, fixed price
Assessment
Interviews with R&D, quality and IT; measurement of the real cost of 2–3 documentation processes; identification and prioritisation of use cases by impact, feasibility and regulatory risk; confidentiality level required per case; roadmap with return estimates. Your team's time: 8–10 hours in total.
The fee is deducted from the pilot if you decide to continue.
02
8–12 weeks · fixed price defined after the assessment
Pilot
Implementation of the prioritised use case on a real, bounded project, with a read-only connector, the validation file prepared in parallel, and metrics agreed before starting. If the numbers don't hold up, the pilot says so clearly: that is what a bounded pilot is for.
03
Optional
Ongoing support
Vulnerability watch, regression packages for model updates, workflow evolution, training.
No hour banks: fixed scope and fixed price at every stage.

We know this problem because we have lived it from the inside

Behind this project there are 18 years developing in-vitro diagnostics instrumentation as a senior system engineer and project manager at leading companies in the sector, plus 4 years in space projects. We have written, reviewed and chased design documentation in regulated environments for an entire career: we know this problem because we have lived it from the inside, not from a generalist AI consultancy. That is why this product is deliberately conservative where it must be — nothing touches your quality system without validation and a human signature — and ambitious where it can be: giving your engineers back the hours that mechanical error-hunting consumes today.

What quality managers usually ask first

The way you validate a method, not a calculator: a narrow intended use, a test bench with known planted defects, and a measurable acceptance criterion. Your current loop already manages non-deterministic elements — human reviewers — through process: qualification, signatures, checklists. The same logic applies here, with one advantage: the test bench is repeatable.
They are different, compatible layers. Copilot is horizontal, and its underlying model updates whenever the vendor decides, which makes version freezing — and therefore validation — impossible. This system brings the frozen version, workflows specific to your QMS, and the physical guarantee that your development documentation never leaves your network.
Responsibilities do not change: the tool sits before your review loop and replaces none of its steps. Your reviewers keep reviewing exactly what they review today; the difference is that documents reach them with the mechanical errors already filtered, and the pilot metrics let you see, with data, that the process stays healthy.
That is the usual case — and it usually happens without authorisation, without version control, and with confidential documentation flowing to consumer services. Part of the assessment's value is sizing that uncontrolled use and replacing it with an official, secure, validated channel.

Let's talk about your case

Tell us briefly about your situation and we will reply within 48 working hours. The first conversation carries no commitment: we are as interested in understanding your documentation burden as you are in solving it.

Response time Within 48 working hours
First conversation No commitment
Email hola@lab-robotics.com
Location Barcelona, Spain
✓ Message received. We will reply within 48 working hours.