Why Do Industries Need Custom AI Tools?

When I first started working in the medical device industry nearly 20 years ago, static analysis tools took the limelight and attention of the medical device industry. This was reflected in a 2007 press article, which highlighted a major investment by the United States Food and Drug Administration (FDA) Center for Devices and Radiological Health (CDRH) in a software forensics laboratory. Brian Fitzgerald from the FDA was quoted at the time, saying, “We’re hoping that by talking quietly about stationary analytical instruments, by encouraging stationary instrument vendors to contact medical device manufacturers, and medical device manufacturers staying on top of their technology, that we can introduce this latest idea that we have.”

I have seen this connection firsthand as I field many sales calls from vendors of static analysis tools. Fortunately, I was already grounded in real-world data, so in 2010, I published a paper at the Embedded Systems Conference to secure custom analytics tool solutions. As a point of interest, the custom solution presented in that paper is still in use today and has found a disproportionate number of software errors compared to its OTS counterparts used to enforce federal coding standards. Now, 15 years later, this topic has come up in the context of custom AI tools, and I find myself compelled to speak again.

Repeating pattern (Now with AI)

Deep collaboration with commercial AI platforms and tools such as Cursor, GitHub Copilot, Windsurf, and various business AI web sites demonstrate the power and potential of these technologies and OTS tools. However, riding the wave of this enthusiasm is the misconception that organizations can simply buy and use these OTS tools and somehow fully realize the transformative potential of AI. While I believe this is generally the case, I will stay on track by addressing the unique challenges that medical device manufacturers face. Intuition alone would seem sufficient to support the argument that pre-trained LLMs, despite their vast training corpus, lack the background clarity, regulatory awareness, and access to data required to provide comprehensive information in critical security situations. However, presenting the case for using custom tools requires some logical thinking.

Data Integration

The most important limitation of OTS AI solutions is their inability to access and use proprietary data of an organization or a specific domain. Therefore, Retrieval-Augmented Generation (RAG) architectures, as described, address this limitation by combining the reasoning power of LLM with domain-specific knowledge. The performance of RAG systems compared to pre-trained baseline LLM models on domain-specific tasks was documented, yielding a 30-50% improvement in LLM response accuracy. Custom AI tools can use RAG programs differently:

• Index domain information using semantic embedding
• Retrieve relevant contextual information from these embeddable data sources
• Low LLM answers to domain data
• Maintain organizational security boundaries

Domain Specific Workflow and Process Integration

The FDA’s Quality System Regulation (QSR) and international standards such as ISO 13485 define specific workflows and defer to other standards such as ISO 14971 for risk management and IEC 62304 for software lifecycle processes. This includes authentication and validation functions, change control, and configuration management, etc. Although this information is in the public domain and is part of the larger training corpus available to LLMs, each medical device manufacturer has its own unique quality system based on these standards and principles. What does this mean in practice?

Modern AI tool development uses multi-agent architectures where specialized LLM agents control specific stages of the workflow. For medical device development, this may include:

• Extraction and validation requirements for internal identity specifications
• Analyze projects against regulatory standards, best practices, and organizational domain boundaries
• Generating compliant code following the organization’s coding standards
• Creating test cases for traceability validation on existing documents outside of the immediate LLM context
• Producing documents in appropriate formatting, such as organizational templates

OTS solutions can only provide this level of expertise if they have knowledge of the organization’s processes and their respective quality management systems.

Research shows that LLMs do best with the right tools. The Model Context Protocol (MCP), introduced by Anthropic in 2024, leads the way by providing a universal protocol for connecting LLMs to data sources and tools through a client-server architecture.

Although this is a global standardization effort, the MCP actually reinforces the need for custom tool development rather than eliminating it. Organizations must still build custom MCP servers that understand their domain-specific data structures, implement security access controls, and handle proprietary data file formats. This includes:

• Creating connectors to legacy systems
• Reformats data for MCP resources
• Manage authentication and authorization
• Understanding how to properly disclose data on MCP services
• Proficient in the use of MCP tools
• Maintaining MCP servers as needs change

Cost Effectiveness and ROI

Internal data supports the claim that custom AI solutions outperform OTS options. Therefore, organizations that achieve significant ROI share common characteristics such as deep integration with core business processes, data-driven approaches that gain proprietary information, continuous improvement cycles, and customized solutions that fit specific needs. In addition, custom tool development, although requiring an upfront investment, offers long-term cost benefits such as:

• Unlimited internal use
• Full control over infrastructure and scaling
• Reusable components across multiple applications

Complaints that emphasize the focus of the main product of the organization and are quick to recommend only OTS solutions or the development of the deployment of resources to consultants or vendors with internal resources is the risk of a lack of basic understanding of the nature of the development of AI tools and the strategic value of domain experts. Given the exposure to problem solving, understanding algorithms and data structures, etc., it would not be a stretch to conclude that these transferable skills would support the claim that software engineers with strong foundations can gain expertise in LLM application development much faster than domain experts who can acquire deep technical knowledge of complex systems. Therefore, the dream scenario for an organization wishing to increase the use of AI would be domain experts who are skilled software engineers. The obvious challenge is the appropriate allocation of those resources.

The conclusion

There is a lot of evidence to support the need for custom AI tool development in regulated industries such as medical device manufacturing. While OTS AI solutions can provide value, the future of AI technology in regulated industries will require building intelligent systems that deeply understand and complement domain-specific technologies. AI is rapidly becoming an engineering skill. Organizations that treat this technology as something to be outsourced must re-evaluate their strategic awareness or risk losing competitive advantage.

References

• Chloe Taft. (2007, October). CDRH Software Forensics Lab: Applying Rocket Science to Device Analysis. Medical Devices Today.
• Rigdon, G. (2010, July). Considerations for static analysis of Medical Device Firmware. Proceedings of the Embedded Systems Conference.
• Lewis, P., et al. (2020). Retrieval-Improved Generation of Informative NLP Tasks. Advances in Neural Information Processing Systems, 33, 9459-9474.
• Gao, Y., et al. (2023). Retrieval-Enhanced Generation of Large Language Models: A Survey. arXiv preprint arXiv:2312.10997.
• Park, JS, et al. (2023). Generative Agents: Interactive Simulacra of Human Behavior. arXiv preprint arXiv:2304.03442.
• Schick, T., et al. (2023). Toolformer: Self-Teaching Language Models Using Tools. arXiv preprint arXiv:2302.04761.
• Markovic, D. (2025). Why Custom AI Solutions Beat Off-the-Shelf Options. In the middle.