Join us for Part 3 of our ‘Safety: From Start to Scale’ series for a special Q&A with the creator of the technology behind SafetyGPT, BJ Yurkovich!
Contact: Betsy Barry
Communication Manager
706.206.7271
betsy.barry@acculonenergy.com
Last week, we announced our partnership with SafetyGPT, a groundbreaking GenAI application designed to address functional safety and product certification. Acculon is currently using it for energy storage, electrification, and application-specific guidelines surrounding lithium-ion batteries, adding another layer of safety to our product development processes, as safety is one of our core values. Plus, the creators of SafetyGPT specialize in enterprise innovation, so they let us here at Acculon pick the name of the new GenAI application: we called it SafetyGPT!
Today, we’re excited to have the opportunity to speak with the creator of the technology behind SafetyGPT, who is none other than our own CTO, BJ Yurkovich.
Interviewer: To start, could you share what inspired you to develop SafetyGPT?
BJ: Having a background in a lot of the hype tech spaces, I often find myself doing artificially intelligent things in various industries, some of them being batteries, some of them being in IoT, and others in GenAI. I’ve always been interested in tying systems together across multiple domains–SafetyGPT is just the latest crazy idea that happened to turn into something great. Having worked closely with some of the top teams in the world in autonomous robotics, hyper-scale systems, and energy storage, where safety is paramount, we recognized a huge gap between product engineering and safety engineering and the time it takes to make great, complex products safe. SafetyGPT is simply a tool to help make system safety development more efficient in the product development lifecycle.
Interviewer: Can you explain how SafetyGPT works for those who are not familiar with generative artificial intelligence (GenAI) and how it can revolutionize functional safety within the energy storage industry?
BJ: Imagine you are a new engineer right out of school and your boss comes in and says “We need to certify our new innovative battery product. All our principals are focused on product delivery, so we need you to figure out a certification strategy.” A good engineer may start googling, and find very quickly that certifying a product relies on standards. Reading these standards (after buying them) requires a significant understanding of the engineering, and the product (which is a massive experience-required undertaking by itself) and often results in months of research and coming up with a tight regulatory strategy, even for the most experienced engineers. This is mostly because standards and regulations are written by committees, and written in such a way that they are obtuse so that they (rightly) do not prescribe a solution, which is important for product innovation. These standards are difficult to read, take hours to sort through, and rarely does a product only need to consider a single standard–a good regulatory strategy considers many different standards and requires a clear process and written framework to be submitted to regulators. This is a daunting task for any engineer (or team!) and even more daunting to a young, inexperienced engineer. SafetyGPT makes this process easier.
SafetyGPT eliminates the excessively time-consuming and inefficient undertaking of reading the standards by automatically summarizing them, making the connections with related standards (and your product, if you give the application that info), and then helping to generate a concise, structured regulatory submission that is expertly tailored to your new product. It also increases the quality of energy storage product safety–something that the (lithium) energy storage industry so desperately needs.
Interviewer: Can you give us an example of how an engineer designing a battery for an electric riding mower, for example, might benefit from using SafetyGPT?
BJ: Before SafetyGPT, coming up with a certification strategy for an electric mower would be a daunting task, especially if it has autonomous features (as most tech has nowadays). You might see if you can google what certifications mowers need, only to find a lot of conflicting information about certifying lawn equipment (not electric), consumer equipment (if the mower is consumer grade), mobility standards, and a whole host of EV related standards that involve batteries. Starting super simple: is a mower with batteries treated like an electric bike (UL2271?) or an industrial piece of equipment (UL2580?)? If you ask around, you will get various answers–and answers that end with “…but it depends” qualifier. And that example with some safety standard references is just related to the battery! Now you have to add autonomous considerations, and the standard lawn mower considerations…and you have not even started putting together the necessary documentation for certifications yet. Very quickly, any engineer is looking around and asking the question “Who can I talk to about this that knows something about this!?” Unfortunately, the answer to that question is often “pay high consulting rates for a temporary resource to learn about your product and offer strategic advice.”
“Today, doing functional safety in the energy storage well requires a resource with deep experience in not only functional safety but also deep knowledge of the product & industry.
I see SafetyGPT democratizing functional safety, enabling product engineers to do safety as a core part of their design process–not as a painful afterthought once the product is designed.”
– BJ Yurkovich,
SafetyGPT Creator
& Acculon CTO
BJ: With SafetyGPT, all I have to do is type in the search bar “Tell me about all the standards that relate to an electric riding lawn mower with a lithium-ion battery pack.” From there, I have the opportunity to ask general questions about the application, and SafetyGPT will suggest relevant standards which allows me to select the appropriate ones, ask scoped questions specifically about just that standard, and then add them to my project and regulatory strategy if they seem relevant. It’s almost like having that expensive resource for a fraction of the time and cost (and having that resource be around in 5 years when all the original designers are long gone!).
Once I have built up that standard list that I think is relevant, I can ask SafetyGPT to start a HARA (Hazard and Risk Assessment) and generate some starter documents with contextually relevant safety information that can feed into a functional safety concept, FMEA, and ultimately a test verification plan and report.
This whole process saves hundreds of hours of excruciatingly detailed reading and writing, and provides an amazing force multiplier to an engineering team by having a knowledgeable functional safety consultant on staff who never gets tired (or grumpy, as most FS engineers get after years in the field)!
Interviewer: How does SafetyGPT stay up-to-date with evolving safety regulations and guidelines in the energy storage sector, especially concerning lithium-ion batteries?
BJ: The SafetyGPT team is comprised of some of the leading machine learning (ML) engineers in the applied ML space who also have deep experience in functional safety from automotive, batteries, and aerospace. We are constantly updating standards and fine-tuning to provide up-to-date information in the models. Because a number of them are working on batteries for fun anyway (and are involved in many of the lithium standards committees), SafetyGPT gets some extra-special love in that area.
Interviewer: How does SafetyGPT ensure the accuracy of the information it delivers to users?
BJ: Left unchecked, Large Language Models (LLMs) can hallucinate (generate erroneous answers) hellaciously. We counteracted this by pairing a set of multi-model safety-content fine-tuned LLMs, a knowledge graph that maps standards and content, and a vector database with significant metadata-tuning. The SafetyGPT magic is really in how we tune all of these to provide a grounded, reproducible, and amazingly efficient system for regulatory strategy management. Another great feature is that the models can be deployed on a customer’s (like Acculon’s) infrastructure or cloud service provider (GCP, AWS, or Azure) so that no sensitive data leaves their firewall and they retain ownership of their data. That’s a pretty important aspect, and why ChatGPT, while a great tool, is impossible to realistically use for sensitive applications like safety, legal, and healthcare.
Interviewer: In what ways do you see SafetyGPT shaping the future of functional safety in the energy storage industry?
BJ: Today, doing functional safety in the energy storage well requires a resource with deep experience in not only functional safety but also deep knowledge of the product and industry. Being a relatively new industry, that talent is few and far between. I see SafetyGPT democratizing functional safety, enabling product engineers to do safety as a core part of their design process–not as a painful afterthought once the product is designed. Good engineers love efficiency and SafetyGPT gives engineers that instant dopamine hit that awesome efficiency gains provide, which immediately will result in safer products. Safety never felt so good!
Interviewer: What are your future plans for this application, and how do you see it evolving?
BJ: A number of certifying bodies are already playing around with SafetyGPT internally. So if you go to them and get consulting in batteries, you may be getting some SafetyGPT-backed answers and not know it!
Also, if you are an Acculon Energy customer, you are definitely getting SafetyGPT-informed designs. You didn’t hear it from me, but rumor has it that if you treat the Acculon engineers extra nicely (treats and tips seem to work the best), they may let you play with it yourself!
SafetyGPT has been growing massively fast, so it will be fun to see how it ends up!
Interviewer: Thank you, BJ for sharing your insights about the exciting foray of GenAI into the electrification sphere. I’m incredibly excited about the possibilities that SafetyGPT brings to Acculon and the energy storage industry at large, and I look forward to its continued evolution and impact!