Inside Our Portfolio: Q&A with Aleph Alpha's Jonas Andrulis

Welcome to the latest edition of our series, "Inside Our Portfolio", where we engage in insightful conversations with leaders from our portfolio companies. This series aims to provide you with firsthand perspectives from those at the forefront of innovation and growth.

In this issue, Paul Glaser (Head of Hewlett Packard Pathfinder and Investor Relations) interviews Jonas Andrulis, Founder and CEO of Aleph Alpha, a Germany-based startup at the forefront of the European GenAI market. Aleph Alpha is pioneering AI solutions that prioritize data sovereignty, security, and regulatory compliance, and is partnering with HPE to deliver these cutting-edge technologies to enterprises.

PG: What does Aleph Alpha do?

JA: Aleph Alpha is an AI R&D startup founded in 2019 by industry veterans from Apple and Deloitte with the goal of bringing a new generation of capabilities to complex and critical problems in industry and government. We specialize in the kind of problems where human responsibility is paramount and the relevance of data, IP and value requires high priority for sovereignty, safety and control. For our customers in finance, manufacturing, government and other core industries, we built technology that creates a foundation and drives innovation for the transformation into a new era of knowledge work.

PG: What are your priorities today?

JA: Since closing our $500 million funding almost a year ago, we were able to assemble a phenomenal team of executives and talent in research, product and customer-focused roles. The focus of our teams is now on innovation and opportunities in specific industries. With solid funding, revenues orders of magnitude higher, and our leadership team almost complete, we are now kicking our internationalization into gear. With this trajectory, I feel I’ve come closer to the advice a brilliant and incredibly accomplished entrepreneur once gave me: “It is the job of the CEO to do absolutely nothing.”

PG: What sets Aleph Alpha’s Pharia models apart from others?

JA: When we launched our first LLM model generation, GPT-4 wasn’t out yet, and it was difficult for our customers to use and leverage GenAI in a sovereign and customizable way. This has changed: For general purpose tasks, there are many powerful open- and closed-source models around, and the challenge has become more about the orchestration, customization, scaling, and evaluation of complex AI workflows in an enterprise-compatible way. We’ve focused our research on the kind of problems these open-source models fail in and, for example, developed an architecture innovation that replaces tokenization. Thus, our proprietary tech is enabling the current model generation to excel in the specialization on out-of-distribution languages and unique knowledge better than any other model. We also added some key innovations, such as our explainability and human-in-the-loop interfaces that we developed, to open-source models in our environment. So, with the Pharia stack, our customers combine highly specialized models with standard LLMs into solutions that reinvent the information-based value chains they need to maintain their sovereignty and that are core to their strategic position.

PG: Can you provide examples or case studies where customers have significantly benefited from using Aleph Alpha?

JA: One of the most advanced examples is the use of our technology for a new generation of compliance work. With PhariaCatch we were able to extract the detailed and specific knowledge and relevant experience of expert lawyers in an efficient way and store it in our systems. With this specialization, we are then able to understand and evaluate the latest compliance standards that are about to go live in 2025, check existing documents, and identify areas that need to be changed to comply with the new regulations. The human-machine collaboration here is not limited to a sequential human approval of AI-generated results. When an AI application recognizes that it lacks sufficient confidence in addressing a specific problem, a proactive expert engagement can help the system to auto-learn in a smart way. This approach to human-machine collaboration, which leverages the strengths of each party and learns dynamically, is the foundation of a new kind of knowledge work. Sometimes chatbots are a good choice of tool but in cases where there’s no easy answers a more advanced approach is needed.

PG: What advantages does your AI cluster provide in terms of scalability and performance for enterprise customers, especially in heavily regulated sectors?

JA: Because we work with the most critical problems and are a partner for extremely demanding environments, we built a secure data center some years ago. Some of our work with government and security experts, as well as with health and financial data and the recently launched audit solution, would not have been possible without the guarantees for infrastructure set up and data handling. Our cluster and infrastructure team also allow us to drive innovation down to the accelerator level. The joint research we conduct with Graphcore, Qualcomm, and AMD ensures that we can take advantage of different hardware, and our customers have full sovereignty in their choice of chip. Through the work of our infrastructure team, we will also become the first European installation of Cerebras’ brilliant CS-3 chip.

PG: How do you see the evolving EU regulatory landscape impacting the development of trustworthy and independent AI solutions in the future?

JA: I am pleased that the Artificial Intelligence Act resulted in a document that makes sense in many of its covered aspects. Many of its details and the effects on standardization are just now starting to be negotiated and it will take some time until all the questions have been answered. While established legal standards help by removing uncertainty, the entire process has diverted significant creative energy from innovation to compliance. We are at the beginning of a new era, and the best innovators will build empires in the following years. AI constitutes perhaps the biggest economic opportunity in tech ever, not just for AI companies, but for every industry. A complex and expensive regulatory standard can lead enterprises to focus primarily on compliance and choose to adopt technology developed by others. With this strategy, where the enterprise is mainly a paying user, it should not be a surprise that the value potential will be captured by others. I believe that for the huge challenges of this generation — demography, climate, fairness, geopolitical tension —innovation is the only feasible path to success, and decisive action is needed. Europe will look different in ten years; with innovation, we can ensure a future with prosperity and stability.

If you like this article consider subscribing to our bi-monthly newsletter to get information about our portfolio, solutions, and insights delivered to your inbox.