Research
The next generation of enterprise agents
starts with structure.
Traize Research explores how a structured World Model of your organisation can power smaller, cheaper and more capable agents.
Graph intelligence, knowledge engineering, agent systems and small language models.
Coming soon
Small agents,
grounded in the World Model.
What if enterprise agents did not have to reconstruct organisational context from scratch?
We are researching compact language models that learn to navigate and act on the Traize World Model, using structured knowledge to reduce repeated reasoning. Build the knowledge layer continuously, reuse it across every task.
- Observe
- Navigate
- Evaluate
- Act
Task
What is blocking Project Atlas?
Grounded answer
Project Atlas is waiting for the security review before access.
Evidence
- Access policy, version 3
- Mail from IT security
Small models in structured environments, trained to find and use the right knowledge. Conceptual illustration, not an actual training run.
Conventional approach
- Retrieve
- Summarise
- Rebuild context
- Reason
Repeated context assembly across tasks.
Traize Research approach
- Living World Model
- Small agent
- Targeted reasoning
Reusable structure intended to reduce repeated inference work.
Our goal: frontier level utility for specific enterprise tasks at a fraction of the inference cost.
Less repeated contextSmaller modelsLower inference cost potential
What we are working on
Four research directions,
one World Model underneath.
World Models for Enterprise AI
Structured representations that make organisational knowledge usable for intelligent systems.
Small Language Models
Compact models trained to reason efficiently when the right structure is available upfront.
Agent Systems
Agents that navigate, query and act inside the World Model instead of raw text.
Graph Intelligence
Algorithms for critical knowledge, freshness, dependencies and organisational structure.
Our team
Built by researchers
who ship.
Research experience in machine learning, language models and secure systems, tested against real organisational complexity.
Lukas Bierling
CTO
World models, reinforcement learning and action conditioned video models.
Retrieval, knowledge graphs and model training. M.Sc. Artificial Intelligence, Amsterdam.
LinkedIn
Adrian Henkel
CPO
Generative AI, language model integration and enterprise AI systems.
Generative AI, language model integration and model training. M.Sc. Bioinformatics, TU Munich.
LinkedIn
Dr. Johannes Köstler
CISO
Security and data sovereignty for knowledge systems.
Security and data sovereignty. PhD in IT security, Passau.
LinkedIn
The feasibility study for Traize is funded by the German Federal Ministry for Economic Affairs and Energy.
Research notes
Notes from our research,
on world models, agents and efficient AI.
Technical notes, experiments and perspectives on what new AI capabilities mean for enterprises.
- AgentsWhy enterprise agents need structure before reasoningIn preparation
- World ModelsFrom unstructured data to world modelsIn preparation
- Small Language ModelsCan smaller models win with better context?In preparation
- Graph IntelligenceGraph intelligence for organisational knowledgeIn preparation
- Enterprise AIWhy inference cost matters more than most teams thinkIn preparation
Follow our research