FROM POC TO PRODUCTION
Explainable AI & measurable KPIs
Robust governance
Artificial Intelligence
We do not build AI as an isolated demo, but as a deployable solution: relevant to the use case, technically robust and properly documented for regulated environments. The result is solutions that not only work, but also remain explainable, verifiable and economically sound.
Clients & Industries
We deliver AI solutions for regulated and highly dynamic domains — where quality, technical robustness and explainable decisions all matter at the same time.
Our AI products
Our product building blocks complement project work wherever structured assessment, robust evidence and reusable workflows are essential.
Thetis
Thetis makes AI systems systematically assessable. The platform captures systems, classifies risks, generates robust assessment results and makes them usable for governance, audit and management.
Especially in regulated environments, strong model performance alone is not enough; you also need clear evidence of which requirements are met, where the risks lie and how technical results can be classified from a regulatory perspective.
✔ Structured assessment of models and AI systems
✔ Clear evidence for governance and documentation
✔ Bridge between technical implementation and regulatory classification
Whyond
Using modern causal analysis methods, we identify more of the decisive influencing factors in defect and root-cause analyses. In finance, the same methods help evaluate audience targeting and actions based on their actual impact.
Whyond turns this approach into a solution for corner-case search, root-cause analysis and regulatorily robust evidence — wherever overlooked quality issues need to be made visible and data collection should be targeted more precisely.
✔ Make corner cases and overlooked quality issues visible in a targeted way
✔ Identify causes instead of mere correlations
✔ Combine knowledge and data in a causal model
Capabilities
From data and model logic to KPIs, assurance, governance and integration into existing product and process landscapes.
We use causal models to better understand relationships and prioritise measures more effectively. This is especially valuable where pure correlations are not enough — for example in root-cause analysis, targeting, policy optimisation or the evaluation of technical influencing factors.
For systems that need to make decisions in dynamic environments, we develop RL-based strategies that combine performance, stability and safety requirements. We pay particular attention to simulation, testability and a sensible operating model.
We design agentic workflows and domain-specific language models wherever complex tasks need to be broken down into sub-steps, combined with rules and executed in a traceable way. Boundaries, roles and observability are key for us.
We help teams understand substantive requirements, build technical evidence and align their solution architecture so that governance, risk management and documentation fit together.
AI must not only work; it must also behave reliably under realistic boundary conditions. That is why we treat data quality, robustness, edge cases, monitoring and traceable release criteria as an integral part of the development path.
From data understanding to evaluation, we build the foundation for data to be interpreted in a professionally meaningful way, visualised clearly and compared effectively. Clean data work, reproducible pipelines and robust operating models are also the basis for any AI project. We connect feature engineering with evaluation into a coherent, client-ready overall picture.
Case Studies
The following projects show three different perspectives on our portfolio: production-oriented driving functions, AI built from existing vehicle data, and a product for governance and compliance.
#AI Driving Guardrails
We are developing a system for highly automated driving functions that generates emergency trajectories designed to avoid collisions while complying with predefined safety objectives.
The focus is on an AI-supported decision logic that derives robust evasive and braking manoeuvres from the current driving context and can be described clearly for later evidence.
#Digital Sensor Twin
Instead of adding extra hardware, existing vehicle telemetry is used to detect driver steering wheel contact in software. This reduces the cost of expensive sensing and creates a scalable solution for increasingly complex architectures.
The challenge lies in deriving robust conclusions from real signals: preparing data, identifying suitable features, training a model and evaluating the results so that they can inform development and product decisions.
#Thetis
Thetis turns scattered requirements into a structured process: capturing AI systems, classifying risks, generating assessments and making the results usable for governance, audit and management.
Explainable AI & Governance
Governance is not an add-on at the end for us. It belongs in development, in assessment and in communication with everyone who carries responsibility.
We are familiar with the substantive requirements and stay up to date with current and upcoming regulatory frameworks. This enables solutions to be built not only for performance, but also for compliance.
AI is not a black box for us. We develop representations and tools that make decisions understandable, assessable and usable for people in day-to-day project work.
We define evaluation metrics so that domain expertise, technical quality and governance speak the same language. That makes progress and risk tangible.
