February 2025 Summaries

Generative AI has gone viral with the release of OpenAI's GPT-3 model, which is now accessible to hundreds of millions of people through an application. However, going beyond models and building successful products requires understanding four key factors: UI/UX design, a complex Compound AI system, monitoring, and effective use of generative models. A good UI makes raw capabilities accessible while guiding users in meaningful ways, while the underlying system should be updated regularly to ensure performance and cost-effectiveness. Users interact with a Compound AI system consisting of multiple components, such as data storage and ML models, but do not need to know the details; instead, they care about usability and usefulness. Effective monitoring is crucial to gain insight into system behavior and generate valuable data for analysis. By focusing on turning AI capabilities into user value, organizations can build successful products that go beyond just using the latest model.

deepset, recognized as a leading AI startup in Germany, is building fast, secure, and accurate Generative AI (GenAI) applications through a partnership with AWS. The company, led by CEO Milos Rusic, is leveraging this collaboration to accelerate enterprise adoption of domain-specific sovereign AI, particularly with the integration of Meta's Llama Stack. As a result, deepset has gained attention from prominent organizations such as Airbus, The Economist, and OakNorth, and has been featured on Sifted's 2025 "Rising 100" list for B2B SaaS companies.

In an article by Milos Rusic from the Forbes Technology Council, the discussion centers on the evolution of AI systems beyond their current limits, emphasizing the importance of smarter configurations known as Compound AI, which integrate components like Agents and Retrieval-Augmented Generation (RAG) for enhanced functionality. Meanwhile, deepset, an AI company, has been recognized as a leading AI startup in Germany by WirtschaftsWoche and is actively advancing the adoption of domain-specific sovereign AI through its collaboration with Meta's Llama Stack, aiming to benefit enterprises, the public sector, and defense. Deepset's innovations have also earned it a spot on Sifted's "Rising 100" list for B2B SaaS, with notable clients including Airbus, The Economist, and OakNorth.

deepset is gaining recognition as a leading AI startup in Germany, highlighted by its inclusion in WirtschaftsWoche's list of top AI startups and its rise on Sifted's "Rising 100" for B2B SaaS in 2025. The company is accelerating the enterprise adoption of domain-specific, sovereign AI by collaborating with Meta's Llama Stack, aiming to make this technology applicable to various sectors including public and defense. deepset's influence is further underscored by its partnerships with organizations such as Airbus, The Economist, and OakNorth, who utilize the deepset AI platform to enhance their operations. An interview with deepset's CTO, Malte Pietsch, discusses the implications of the European AI Act for AI companies in Europe, reflecting the company's active engagement with regulatory developments in the AI landscape.

Large language models (LLMs) have matured enough to power autonomous AI agents that can understand nuanced context, operate various tools with minimal human intervention, and perform multi-step tasks. Unlike other generative AI applications, these agents actively pursue goals and decide on the right tools to achieve them. They combine several key mechanisms to tackle complex tasks effectively, including understanding a situation, weighing options, and choosing the best path forward. Agents use well-defined tools and APIs, clear instructions, memory, and continuous evaluation to accomplish specific tasks. The modular nature of compound AI systems enables teams to start simple and incrementally expand system capabilities, making practical decisions about system architecture as they evolve. Evaluating agents requires tracking outcomes and process efficiency, and the deepset AI Platform empowers users to harness this new class of AI by rapidly prototyping agentic solutions and iterating based on real-world feedback.