Jans Aasman of Franz Inc. on Building AI in Production
As CEO, I lead the vision and development of AllegroGraph, a Neuro-Symbolic AI platform that guides enterprise customers in building agentic AI applications that reason over complex data and operationalize trustworthy, explainable AI at scale.
Please introduce yourself and tell us what you are currently building, deploying, or responsible for.
I am the CEO of Franz Inc., where I work closely with enterprise customers to help them design and deploy agentic AI systems that can operate reliably in real-world environments. At the core of our work is AllegroGraph, a Neuro-symbolic AI platform, which uses semantic knowledge graphs to unify fragmented data and provide a structured foundation for reasoning. By combining this with machine learning, ensemble LLMs and natural language interfaces that allow users to “talk to their data,” through an interface called GraphTalker, we enable organizations to move from disconnected insights to continuous, context-aware decision-making. Our focus is on making AI practical in production - ensuring it is explainable, verifiable, and capable of supporting high-stakes operational decisions at scale.
What kind of AI systems, workflows, or use cases are you closest to right now?
Enterprise AI systems that integrate semantic knowledge graphs, natural language interfaces, and reasoning engines - particularly for use cases involving complex data integration, decision support, and agentic workflows across structured and unstructured data environments.
What real problem are you solving, for whom, and under what constraints?
Our customers span government, healthcare, finance, life sciences and logistics, and they need AI systems that are explainable, auditable, and reliable under strict regulatory, operational, and data integrity constraints.
Where have you seen measurable operational impact from AI so far, and where has the reality fallen short of expectations?
We’ve seen strong results from agentic AI applications in areas such as cost reduction and spend management, where these systems are already delivering measurable business value. In healthcare, our semantic knowledge graph implementations place the patient at the center of a unique event-based model that integrates all relevant data with clinical knowledge to drive 360-degree decision-making. However, expectations around fully autonomous AI have outpaced reality. Most organizations still require systems that combine automation with human oversight, especially where accuracy and trust are critical.
What have you learned about making AI work inside real organizations, products, or operating environments?
One of the most important lessons is that AI only works in real organizations when it is grounded in how data actually exists: fragmented, distributed, and often inconsistent across systems. Enterprises cannot rely on siloed datasets or isolated models - they need a unifying layer. This is where semantic knowledge graphs become critical, as they connect ALL data sources into a coherent, contextual model that reflects real-world relationships and meaning.
Equally important is accessibility. We have developed a natural language interface called GraphTalker that allows users to “talk to the data” and removes the barrier between complex data systems and users. When combined, knowledge graphs and natural language interfaces create a foundation for AI that is not only more powerful, but also explainable and aligned with how organizations actually operate.
What is one prediction you have about how production AI will evolve in practice over the next 12 to 24 months?
Agentic AI systems need a trusted orchestrator. I believe we will see a shift toward architectures where semantic knowledge graphs - powered by neuro-symbolic AI - serve as the cognitive operating system for agentic applications. As organizations move beyond isolated LLM deployments, they will need a persistent, structured layer that connects data, domain knowledge, and supports reasoning across complex workflows.
Semantic knowledge graphs can provide this foundation by grounding AI in a real-world context, while neuro-symbolic approaches combine statistical learning with symbolic reasoning to deliver more accurate and explainable outcomes. LLMs become interfaces and tools within a broader system, while the knowledge graph orchestrates context, memory, and logic.
