
Multi-agent AI software and algorithm validation for healthcare
We design, build and validate multi-agent systems under IEC 62304, EU AI Act, ISO 42001 and data protection (GDPR and Spain's LOPDGDD), with full decision traceability. Based in Valencia, Spain, we work in Spanish and English.
What do we do with multi-agent AI systems in healthcare?
- Multi-agent architecture
- Agent orchestration with tool use, memory and human oversight.
- Evaluation and benchmarking
- Domain-specific evaluation design, metrics and reproducibility.
- Algorithm reliability and fairness
- We verify the system performs consistently across all patient profiles and detect when it loses accuracy over time.
- IEC 62304 validation
- Software classification, requirements, verification, validation and traceability.
- EU AI Act audit
- AI risk management system, data governance, transparency and logging.
- Operational traceability
- We record every decision made by the AI system so you can demonstrate what it did, when and why — essential for inspections and post-market follow-up.
- ISO 42001 · AI Management System
- Implementation of the AI management system per ISO 42001: policy, roles, AI risk management and continual improvement.
- Health data · EHDS and HIPAA
- Designing AI systems that handle health data under European Health Data Space requirements and HIPAA protocols: access control, traceability and data minimisation.
What does a reference multi-agent architecture look like?
An orchestrator splits the work between agents that each do one job. A person reviews the result and every step is recorded.
- Orchestrator
- Receives the request, decides which agent acts and in what order.
- Specialist agents
- Source retrieval, reasoning, domain tools, validation and report drafting.
- Human review
- A qualified person approves or corrects before anything is released.
- Audit log
- Every input, decision and output is stored, so you can reconstruct what the system did and why.
FAQ
What is a multi-agent AI system?
A multi-agent AI system splits a task among several specialised agents, coordinated by an orchestrator, that use controlled tools. In healthcare we design it with a record of every decision, so what each agent did and why can be reviewed.
How is a health AI algorithm validated?
A health AI algorithm is validated by comparing its performance against an independent reference dataset, with metrics defined in advance, and by analysing its robustness and its fairness across subgroups. We document the protocol and the results so they are reproducible.
Does IEC 62304 apply to software with AI?
IEC 62304 applies to the life cycle of software that is a medical device or part of one, with or without AI. It organises requirements, architecture, verification and maintenance. We use it as the base and add the model-specific evidence.
What is human oversight in an AI system?
Human oversight is the set of measures that let a person understand, monitor and, if needed, stop or correct an AI system. We design it into the system: what is shown, who decides and what is recorded.
Do you certify AI systems under ISO 42001 or the AI Act?
No. We design, document and validate AI systems and prepare the evidence. Certification and conformity assessment are carried out by accredited or notified bodies, not by us. We prepare and guide; we do not certify.

Every medical product has its own path to market.
What stage are you at? We help you map the next step: regulatory, clinical trial or software validation.
Illustrative image generated with AI