FoundMed

Services

From your own AI subsystem to full system validation

Six services that run in sequence. Start at the step you need; each one produces the evidence the next step relies on.

01 · Build

Building your own AI subsystem

You know where AI could help: a summary, a risk score, a classification step, a document check. We help you scope it, prepare the data, and build it as a defined subsystem inside your own process, owned by your team.

We fix the intended use and its limits first, so the component is designed from day one to be validated.

Typical deliverables

  • Intended-use statement and scope
  • Data inventory, access and preparation plan
  • Model or rules component built and versioned
  • Interfaces to the surrounding system
  • Handover to your team

02 · Train

Training foundation models

Pretraining and adaptation on your own data. We design the corpus, set the vocabulary for your coding systems, and run pretraining and fine-tuning as a reproducible, versioned process.

The resulting weights come with a provenance record, not just a checkpoint file.

Typical deliverables

  • Corpus design from EHR, claims or registry data
  • Code vocabulary and sequence representation
  • Pretraining and task fine-tuning runs, versioned
  • Model card and training record

03 · Measure

Evaluation and benchmarking

An honest read on whether a model, yours or a vendor's, is good enough for the decision it will support. We set the endpoint and acceptance criteria up front, test on held-out, temporal and external data, and compare against the method you use today.

Typical deliverables

  • Endpoint and pre-set acceptance criteria
  • Held-out, temporal and external validation
  • Subgroup and calibration results
  • Benchmark against the current method
  • Evaluation report usable as validation evidence

04 · Integrate

Workflow assessment and integration

We map the process as it is actually performed, find where an AI component changes the outcome and where it does not, and design the integration into your existing systems and SOPs with clear human review points.

Typical deliverables

  • Current-state process map
  • Opportunities scored on value and feasibility
  • Target-state design with human-in-the-loop steps
  • Integration build and handover

05 · Validate

System validation: CSV and CSA

Risk-based computerized system validation following GAMP® 5 Second Edition, for laboratory instruments, production equipment, spreadsheets, enterprise systems and systems with AI elements.

For AI components we add what models need: data lineage, version control over weights and prompts, performance acceptance criteria, and a monitoring plan.

Typical deliverables

  • GxP impact assessment and GAMP categorization
  • Validation plan and supplier assessment
  • URS, functional and design specifications
  • Risk assessment (GAMP® 5 model and FMEA)
  • Test protocols and reports: functional, system and user acceptance
  • Requirements traceability matrix and validation summary report

06 · Comply

Compliance and maintaining the validated state

Validation is the start of compliance, not the end. We set up the controls and routines that keep a system in a validated state, and we monitor AI components for drift after go-live.

Typical deliverables

  • Data integrity (ALCOA+) controls and audit trail review
  • EU GMP Annex 11 and 21 CFR Part 11 gap assessment
  • Change, configuration and patch management
  • Risk-based periodic review
  • Performance and drift monitoring for AI components

How we engage

One step or the whole range

Most clients start with an assessment, then decide how much of the work to hand over.

Assess

Gap or readiness assessment

A short review of a planned AI component or an existing system against GAMP® 5, Annex 11 and Part 11, with a prioritized plan.

Deliver

Project delivery

We build, evaluate, integrate or validate a defined system with your team, and hand over the documents and evidence.

Sustain

Ongoing compliance support

Periodic reviews, change assessments and monitoring to keep systems in a validated state after go-live.

Next step

Tell us where you are on the range.

Whether you want to build your first AI component or need an existing system validated, send a short description and we will suggest the right starting point.