Skip to main content

Resources

Engineering knowledge for medical device teams. Standards explained, methodologies unpacked, and the thinking behind how MANKAIND works.

Articles

How to write medical device requirements that hold up under FDA review

EARS notation, the six quality attributes FDA reviewers apply, the five most common mistakes, and how to build traceability into requirements from the first line.

Using AI to write medical device requirements — what works and what doesn't

An honest assessment of what general-purpose LLMs get right, where they consistently fail in a regulated context, and what changes when AI is purpose-built for medical device development.

From spreadsheet to graph — a better way to think about traceability

The traceability matrix is what you submit. Traceability is the thing you either have or don't. This guide reframes requirements traceability as a directed graph — and why that changes how you develop.

The medtech CTO's first 90 days — a regulatory decision framework

What to decide in the first 30, 60, and 90 days: device classification, QMS architecture, tool selection, team structure — and the three patterns that cost startups 12 months.

SOUP management for SaMD teams — a practical 2025 guide

IEC 62304 requires every third-party dependency to be treated as a regulated artifact. How to build a SOUP inventory, evaluate risk per item, handle version updates, and manage ML frameworks as SOUP.

How to do a risk analysis for a device that uses AI or machine learning

ISO 14971 applied to AI/ML failure modes: model drift, distribution shift, dataset bias, confidence miscalibration, and how PCCP intersects with post-market risk management.

How to build a quality system before you have a quality team

The minimum viable quality system for a pre-Series A medical device startup — what FDA design controls actually require, stripped of ceremony, and the five elements you cannot defer.

Predetermined Change Control Plans — what FDA wants and how to write one

FDA finalised PCCP guidance in December 2024. The four required components, a worked example for retraining a diagnostic AI, and what makes a PCCP defensible versus vague.

Context engineering for regulated products: why what you feed the agent matters more than the prompt

Context engineering replaced prompt engineering as the dominant AI practitioner framing in 2025-2026. For medtech, the discipline is even more consequential — the four failure modes, context architecture for specific regulatory documents, and what good retrieval design looks like.

Why the reviewer agent is the most important part of your AI documentation stack

Multi-agent orchestration with self-verification is the dominant enterprise AI architecture in 2026. The reviewer-agent pattern is the production answer to LLM hallucination in regulated settings — where hallucination rates on domain-specific queries still run 15–23%.

MCP in medtech: the protocol that lets your AI agent talk to your QMS, ERP, and CAD files

Model Context Protocol (MCP) is the architecture that lets one AI agent read your PLM, QMS, test management system, and FDA database in real time — with every access logged for audit. What it looks like in a regulated environment, including governance, version pinning, and write gates.

The FDA is using agents to review your 510(k). Are you using agents to write it?

The FDA deployed agentic AI agency-wide in January 2026 for premarket reviews, 510(k) processing, and postmarket surveillance. What FDA's internal agents actually do, how hand-assembled submissions fail automated consistency checks, and what a submission-ready engineering record looks like.

Five months to EU AI Act enforcement: what medical device teams need to do before August 2026

EU AI Act full enforcement hits August 2, 2026. Most AI-embedded medical devices face dual conformity under both MDR and the AI Act. The six Annex IV documentation requirements, which MDR artifacts satisfy them, and the four things you cannot defer.

RAG is winning in regulated industries. Here's the architecture that explains why.

In a regulatory submission, every technical claim needs a traceable source. RAG generates citations alongside output by design — fine-tuned models bake stale standards into weights. Corpus design, chunking strategy, hybrid retrieval, and evaluation for regulatory documentation.

IEC 62366 usability engineering — what medical device teams actually need to document

IEC 62366-1 governs the usability engineering process for medical devices. Formative vs summative evaluation, hazard-related use scenarios, the UEF, and the structural linkage to ISO 14971.

FDA medical device cybersecurity under Section 524B — what changed and what you now have to prove

Section 524B made cybersecurity binding for every cyber device. SBOM, threat modeling, post-market vulnerability management, and the distinction between security and safety risk.

ISO 10993 biocompatibility — from evaluation planning to the test battery

ISO 10993 is an evaluation standard now, not a testing checklist. Contact categorization, ISO 10993-18 chemical characterization, and the shift from testing-first to biological evaluation.

MDSAP — the single audit that satisfies five regulators

One audit, five jurisdictions: FDA, Health Canada, TGA, ANVISA, PMDA. The seven-process audit model, the 5-grade scoring system, and what MDSAP does — and does not — replace.

IEC 60601 medical electrical equipment — the standards hierarchy, explained

IEC 60601-1 Edition 3.2, collateral standards, particular standards, essential performance, and the EMC immunity shift in 60601-1-2 Edition 4.1.

Guides & glossary

Use cases

Standards explained