



















“Half of recalls of AI-enabled medical devices come down to design flaws, not manufacturing.” — National Library of Medicine


Standalone SaMD that performs a diagnostic or therapeutic function without being part of a physical device.
Software in a medical device, including firmware, RTOS work, and real-time control on the software side.
Model-driven devices built with a change control plan and monitoring designed in from the start.
DICOM viewers and medical image analysis software that present and interpret medical images for review.
Medical device app development for the mobile and desktop applications that pair with an instrument.
Device integration and connectivity that move readings from the device to cloud and clinical systems.
Backends that ingest, store, and analyze device data at scale under regulated controls.
Re-engineering existing software on aging platforms without losing validation status.
Verification and validation performed on software your own development team built.
The software sections of a submission, assembled from the records the build already produced.
HL7, FHIR, and DICOM interfaces so device output reaches healthcare systems in a usable form.
Secure software architecture, threat modeling, and the SBOM a submission now requires.

“Only 16.7% of machine learning-enabled devices cleared by the FDA include a predetermined change control plan.” — National Library of Medicine






















It is the engineering of software that is part of, or is itself, a regulated medical device, covering software as a medical device and embedded software inside hardware. Because it can affect patient safety, it is built to IEC 62304, ISO 14971, and FDA requirements with verification and validation throughout.
SaMD performs a medical purpose on its own, such as software that analyzes images to flag a condition. SiMD, or embedded software, runs inside a physical device and controls its function, like the firmware in an infusion pump. Both fall under IEC 62304, but embedded work adds real-time constraints.
IEC 62304 defines the software life cycle and assigns safety class A, B, or C, working alongside ISO 14971 for risk management and ISO 13485 for the quality system. In the US, the Quality Management System Regulation now incorporates ISO 13485:2016 by reference in 21 CFR Part 820.
The main types of software are standalone SaMD, embedded firmware, companion and mobile applications, and the cloud and analysis software behind them. Each is regulated according to its role and intended use rather than its platform.
Yes. IEC 62304 does not mandate a lifecycle model, so agile works provided the required activities, records, and traceability are produced. In practice, lifecycle activities map onto sprints and risk and design records update per increment rather than at the end.
AI models are regulated as software under the same lifecycle and risk standards, plus AI-specific expectations around Good Machine Learning Practice and a predetermined change control plan describing how the model may be updated. Data quality, bias, and real-world performance monitoring are part of the development record.
Under Section 524B of the FD&C Act, devices with software must address cybersecurity in premarket submissions or risk refusal to accept. That means a software bill of materials, threat modeling, security risk management, and a plan for postmarket patching and coordinated vulnerability disclosure.
Yes. V&V is built into the lifecycle with traceability from requirements through risk controls to test results, producing the design history file and the software sections of a 510(k), De Novo, or PMA. We support your regulatory team rather than replacing regulatory or legal counsel.
We work inside our clients’ ISO 13485 quality management systems and follow their procedures and document control, which is what keeps their certification and audit trail intact. Ask us directly about our own certifications and we will answer specifically.
Ask for regulated lifecycle experience, V&V and submission track record, embedded and AI capability, and how the team handles evidence. Directory roundups such as a “top 10 medical device software” or “10 medical device software development” list are a starting point, but the useful test is whether a leading medical device software development team can explain its traceability approach in detail.
Yes. We modernize existing software, take over maintenance of medical device software projects, and perform independent V&V on code built elsewhere, including software development for medical devices already on the market.
Yes. Device manufacturers often need Salesforce, service, and quality platforms connected to product data, and we build those integrations alongside the regulated software rather than as a separate project.
Cost depends on safety classification, whether the work is embedded or SaMD, the role of AI, cybersecurity scope, and the submission required. A low-risk class A tool costs far less than a class C AI-enabled device heading for clearance, and a precise estimate follows discovery.
No. We are a software engineering firm and cover the software layer of hardware and software programs, working alongside your electronics and mechanical teams or a specialist hardware partner.