Have a question not answered here? Contact us: we're happy to help.
Our solutions run entirely within your own IT environment or a private cloud managed by you. No data exchange takes place with external parties. You retain full control over where data is stored and processed. AI models are hosted locally, queries are processed within your network, and answers are generated within your environment.
Yes. We integrate with common IAM systems such as Azure AD, Okta, and other SAML/OIDC providers. Access rights from your existing environment are respected: employees only see the documents and data they are already authorised to access. This applies to both search results and AI-generated answers.
We launch an initial pilot within 30 days. A full production implementation typically takes 8 to 14 weeks, depending on the complexity of your IT landscape, the number of data sources to integrate, and the desired use cases. We work with a structured implementation process that minimises risk.
We are model-agnostic but not model-dependent. We deploy both commercially licensed and open models within your infrastructure, selected based on your performance, security, and compliance requirements. Beyond that, we fine-tune existing models on your domain data for superior accuracy, and when no existing model fits, we build and train custom models from scratch. That is the difference between an integrator and a builder.
Every interaction with our AI systems is recorded in audit logs. These logs contain: who submitted the query, which sources were consulted, which answer was generated, and whether human approval was granted. Answers always include source references to underlying documents. A human employee retains final responsibility for decisions.
Our solutions are modular. New use cases can be added incrementally without disrupting the existing implementation. Horizontal scaling happens via your own infrastructure: you set the pace. We advise on capacity planning and can support expansion to new departments or data sources.
For a pilot you need a defined dataset (e.g. a document collection or knowledge base), a secure environment for the AI infrastructure (physical or virtualised), and a project sponsor with authority. We handle the technical setup and guide the process. A pilot has minimal impact on your existing IT operations.
Yes. We have experience with deployments in fully isolated environments without internet connectivity. This requires specific preparation regarding model distribution, update procedures, and monitoring, but is part of our standard service.
Model updates are performed according to a schedule approved by you. Updates are first tested in a staging environment before going to production. We offer various maintenance contracts, ranging from basic support to 24/7 managed services. You determine how much operational responsibility you delegate to us.
Costs depend on the scope: the number of use cases, the complexity of your IT landscape, the number of data sources to integrate, and the desired service level. We work with a fixed project price for the initial implementation and transparent licensing models for ongoing use. After an intake meeting, you will receive a concrete quotation.
Knowledge transfer is an integral part of every implementation. We train your administrators in daily management, users in effective interaction with the AI systems, and document all configuration. The goal is for your organisation to work independently with the solution after completion, with our support as a safety net.
Our architecture is designed to comply with common compliance frameworks. Because the solution runs within your own infrastructure, compliance with sector-specific regulations is largely ensured by your own IT controls. We provide documentation and can facilitate an external auditor if desired.
Yes. This is core to what we do. When off-the-shelf models are insufficient, we design and train custom neural networks from scratch, or fine-tune existing models on your domain data. Our research team works in deep learning, transformer architectures, and representation learning. Whether it is geological classification, document understanding, or predictive maintenance, we build the model that fits your problem rather than forcing your problem to fit an existing model.
Fine-tuning takes an existing pre-trained model and adapts it to your specific domain, terminology, and data. This is faster and cost-effective for most use cases. Building a custom model means designing a new architecture from scratch, trained on your data for your specific task. This is necessary when no existing model can achieve the required performance. We help you determine which approach is right for your use case.