I’m always eager to learn from innovators who have trailblazed a unique position for themselves in the AI space. Steen Graham, founder of Metrum AI, is one of them. Steen leveraged his deep experience developing cutting-edge AI semiconductor chips to launch his own company helping businesses harness the full potential of AI. He’s strongly positioned to help them navigate the complex web of infrastructure choices, model optimizations, and industry-specific applications.
I recently had the pleasure of talking with Steen about solving real-world AI challenges and the future of enterprise AI for the AMD EPYC TechTalk podcast series, available here. You can also find highlights from the interview below.
A lot of enterprises have experimented with AI productivity tools such as chatbots, but there’s a growing demand for AI systems that can drive top-line business outcomes. Companies in sectors such as finance, healthcare, telecommunications and legal services are increasingly looking for high-fidelity AI applications that can transform their operations.
Today's AI landscape requires more than just testing foundational models for accuracy and throughput. The advent of transformer models and generative AI opened new opportunities and challenges in training, fine-tuning and deploying models. Enterprises must now assess the entire ecosystem – including vector databases, embeddings models, and multimodal AI applications – to ensure that AI solutions meet industry-specific requirements.
After 17 years in the semiconductor industry working on purpose-built AI chips and optimization toolkits to make models run faster, Steen got excited about helping solve the complexities of deploying AI workloads at scale. Metrum AI addresses these challenges by building testing tools that allow for rigorous evaluation of AI workloads across millions of permutations, ensuring real-world applicability and performance.
“We want to focus on helping businesses transform beyond productivity tools,” Steen said. “How do we actually impact their key performance indicators as a company – like quality of service or seeing more patients in the healthcare space? We built a really vast portfolio of solutions that orient around business outcomes and driving top level KPIs versus just making people a bit faster and a bit more productive with the chatbot, for example.”
Companies in regulated industries, such as healthcare and finance, particularly need to make sure AI models meet strict accuracy and safety standards to avoid issues like "AI hallucinations," where incorrect or misleading information is generated. To address these concerns, Metrum AI integrates advanced techniques such as software-defined prompting and retrieval-augmented generation, which enhance the reliability and accuracy of AI outputs.
“Where there's a regulatory pressure, there's customers that are really pushing for these solutions,” he said. “There's a huge opportunity to deploy generative AI, ultimately to enrich people's lives.”
A very timely example is insurance companies dealing with massive hurricane damage. There’s regulatory pressure to make sure people get paid from their insurance policies so they can quickly rebuild their houses and recover their lives. But insurance companies have procedures to go through and policy to execute—assuring fraud mitigation but also combating potential bias. Steen’s multimodal AI solutions can assess damage and provide preliminary early warnings on loss events, assess the policies and put out early recommendations, and then bring the rest of the certified professionals in at a much faster rate to drive outcomes.
Choosing the right infrastructure when deploying AI for these use cases is key. As the complexity of AI workloads increases, enterprises need to make informed decisions about memory footprint and compute capabilities, especially when dealing with multimodal AI that requires multiple models working together. Steen said AMD’s Instinct™ MI300X Accelerators have allowed him the capacity to deploy a full stack enterprise solution with a breadth of models and capabilities.
“On our Metrum enterprise product, we have an enterprise offering that's multimodal retrieval augmented generation,” Steen said. “Thanks to GPUs such as the Instinct MI300X from AMD, we're able to stuff in a ton of models into that: The embeddings model, the language model, audio model, and video models as well, all into one GPU system to deploy some of those critical applications.”
As AI continues to reshape industries, Metrum AI is positioned to lead the charge in helping businesses leverage the power of AI to achieve their goals.
“I just want to help empower business leaders that are ambitious about transforming their business and delighting their customers to drive major impacts on their top line and their other key performance indicators as a company,” Steen said. “We've been lucky enough to be brought in on a number of those opportunities, and I’d love to continue to work with those innovators in making this more realistic and delivering on their vision to transform their business.”
