“We're seeing companies move from simply consuming AI to orchestrating it with their own data“

Off-the-shelf AI tools work well for individual productivity, but they rarely address the complex workflows that enterprises depend on in regulated industries. Financial institutions, government agencies, and aerospace companies need AI that can be precisely tuned to their domain, integrated with their data, and deployed under their full control. This is where deepset comes in – with an open-source framework and enterprise platform that allow organizations to orchestrate custom AI solutions for mission-critical processes. 

Milos Rusic is CEO and Co-founder of deepset. Founded in Berlin in 2018, the company has grown into a platform used by thousands of organizations worldwide. At its core is Haystack, an open-source framework for building agentic and RAG applications, complemented by enterprise infrastructure for governance and scale. In 2025, deepset was named one of Germany's Most Promising AI Startups and selected as the AI orchestration layer in the Deutschland-Stack initiative. Furthermore, deepset was one of the first participants of the AI showroom at the recently launched #ai_berlin hub.

In this interview, Milos discusses how deepset enables "AI with purpose" for business-critical workflows, the connection between open source and enterprise adoption, and Berlin's evolution as an AI hub. He explains why trust and control are non-negotiable in sensitive fields, and where he sees the greatest potential as enterprises shift from consuming generic AI to orchestrating it within their own systems.

Hello Milos, deepset helps organizations build custom AI applications for business-critical processes. What does this look like in practice – can you share an example of how a company uses your platform, and what specific value it creates compared to off-the-shelf AI tools?  

Off-the-shelf AI tools work well for individual productivity, but they rarely address the complex, cross-functional workflows and decision-making processes that enterprises depend on. With our Haystack framework and the deepset platform, organizations can orchestrate LLM solutions that are precisely tuned to their industry domain, data, and compliance requirements. For example, financial institutions like OakNorth Bank and credX use deepset to double the speed of their lending due diligence; aerospace companies like Airbus build decision-intelligence systems for the German Armed Forces that enhance safety in defense operations; and government agencies like the BMFTR deploy our technology to to make information accessible for clerks to be able to evaluate grant applications. Solutions like these require transparency, trust, and sovereignty to achieve their intended scale and impact.  

Your open-source framework Haystack has become a widely used standard for building AI applications. How does this open-source foundation connect with your enterprise platform – and what does this combination mean for users in practice?  

Haystack is the foundation for building with AI. It is modular and flexible for innovation, yet stable and reliable for production. Developers at many thousands of organizations worldwide use it to create agentic, RAG (retrieval-augmented generation), and multimodal applications that combine text, speech, audio, and imagery. Take an example like Telus, that just shared their success in building industry-specific agents in agriculture and consumer goods. It lets them prototype quickly, iterate fast, and scale solutions for real-world adoption. They’re one of thousands of organizations that include the biggest names in finance, technology, healthcare, government, as well as small and medium businesses. On top of the open-source framework, these customers are taking advantage of our enterprise-grade support through Haystack Enterprise and deepset’s platform, which make it easy for them to run, scale, and secure these solutions in the cloud or on-premise. For users, that means they can start with open source immediately, get expert help as they grow, and benefit from the infrastructure, governance, and tooling needed to manage and continuously improve many use cases at scale.  

This year, deepset was included in the German AI Startup Landscape 2025 as one of the ‘Most Promising AI Startups in Germany.’ As a company founded in Berlin (incorporated in 2018), what does this recognition say about your journey – and about Berlin’s ability to foster internationally relevant AI companies?  

Being named one of Germany’s most promising AI startups is a great recognition of our team’s focus and talent, and of our mission to help customers build “AI with purpose,” solving complex problems that unlock real, transformational value. It tells me that our journey is on a good track, but more importantly: building true substance in AI and any technological field usually starts before the mainstream is hyping about it. Berlin has played a defining role in this journey. In 2018 we were one of the very few in Berlin to focus on AI – but in a city like Berlin, people are always curious, want to learn and start providing perspectives and input. In the early years, this access to open-minded early adopters and a vibrant tech community where ideas spread quickly was essential, and also because we won our first customers here. The next opportunity for Berlin is scaling up: helping startups evolve into globally impactful companies. I believe we’re seeing that happen, not just with deepset, but with a wave of Berlin-based AI firms that are earning international relevance. The #ai_berlin hub initiative is a great step to foster exchange and build a network that accelerates growth. 

deepset began in Berlin in 2018 and now also operates from New York. What role did Berlin play in the company’s early years, and how do you see the city’s ecosystem today when it comes to supporting AI scale-ups?  

Expanding to New York was a natural parallel step once we saw Haystack’s adoption take off globally, in particular among some of the world’s most respected companies headquartered in the U.S. That global traction has shaped deepset into a true melting pot of talent inside the company too, from Berlin, across Germany, the EU, the UK, and the U.S. — all working toward a shared mission of delivering “AI with purpose” wherever organizations are asking more from AI. Berlin was the perfect place for us to first grow from zero to one, but is also the center of our next scale up phase. The city’s mix of technical talent, creativity, and openness make it ideal for both stages. Today, the ecosystem has matured — there’s more capital, more specialized talent, more demand from established industries, and growing collaboration between public and private initiatives supporting AI scale-ups. Being named as the AI orchestration layer in the Deutschland-Stack initiative shows that Berlin and Germany are building the foundations for sovereign, production-grade AI infrastructure. It’s a sign of how far the ecosystem has come — and we’re proud to call Berlin home as we help shape that next phase.  

Trust, control, and reliability are recurring themes in your positioning. From a broader perspective, why are these qualities particularly important for AI adoption in sensitive fields like finance, government, or healthcare – and how do you address them?  

In sectors like finance, government, or healthcare, mistakes aren’t just inconvenient, they can have legal, financial, or even human consequences. That’s why trust, control, and reliability aren’t buzzwords for us; they’re requirements. We give customers full transparency into how their AI works — step by step — including how answers and actions are generated and how data and models are securely governed. Everything can run where it needs to, whether in the cloud or on-premise, and every component is configurable. That means as your needs evolve, you can adapt your AI without being locked in. This level of transparency and flexibility is what gives enterprises the confidence to deploy Haystack and work with deepset in their most sensitive and mission-critical workflows. This is our definition of “sovereignty” in AI and we see all organizations of a certain size, no matter if governmental or in the private sector, are striving for it.  

Looking ahead, where do you see the greatest potential for enterprise AI – whether in AI agents, enterprise search, or intelligent document processing – and what developments should policymakers, businesses, and society keep an eye on?  

We’re at a tipping point where organizations are learning what it takes to see real results from AI. The shift is clear: we see companies moving from simply consuming AI trained on internet-scale data to orchestrating it with their own data, workflows, guard rails, and goals built in. Traditionally, that level of customization sounds hard or complicated to achieve but that’s exactly what deepset makes fast, simple, and repeatable. The real value comes when these AI architectures are adapted and combined for an organization’s unique environment. That’s where our platform helps to make this level of control readily accessible. For policymakers, this is the moment to support and amplify the Sovereign AI Stack for Germany and our EU neighbors. Businesses should think big and let their best business outcomes lead, focusing on modular AI platforms and practices to adapt AI to their core workflows. For us as a society, the real progress will come from understanding how AI works in business environments as systems, not just as models — that’s how we’ll lead in applying “AI with purpose” to transform work, decision-making, and innovation. 

Thanks for talking to us.