Chatbot for
Machine Data

The assistant uses live and historical data from your IoT platform and supplements this with knowledge from documents, wiki content or intranet pages via Retrieval Augmented Generation (RAG).

The connection is made via APIs or standardized interfaces (e.g. REST) and the existing data access layers of your platform. Adapters are also possible for proprietary or customer-specific systems—via MCP (Model Context Protocol).

You choose the right operating model for your compliance requirements:

• On-Premise / Private Cloud / Hybrid
• SaaS

Individual authentication and authorization, a granular role and rights concept with context control as well as logging and auditing.

The chatbot accepts questions in natural language and accesses live and historical data from the IoT platform as well as information from technical documents (document-based search / RAG) depending on the query. This results in a comprehensible answer, optionally with visualization and recommendation.

RAG ("Retrieval-Augmented Generation") ensures that the assistant not only "freely formulates", but also looks up specific information in your documentation (e.g. service manuals, Confluence, PDFs). Answers are therefore based on specific sources and not on freely formulated assumptions.

Typically, live machine data such as alarms, time series or properties from the IoT platform are connected. Technical documents such as manuals, wikis or best practices can also be used. We will clarify together in the workshop which sources make sense in your case.

• Access to your IoT/data platform (API) for live and historical data.
• Clarified asset context: Which machine/system is meant, which tags/signals belong to it?
• Knowledge base (documents/wiki) plus retrieval layer (RAG) for document knowledge.
• Authentication/authorization, role/context control and logging according to your target architecture.

We integrate the chatbot into existing IoT and data platforms via APIs and standardized interfaces, for example on the basis of Azure, AWS or ThingWorx. We clarify which systems you are using in a brief inventory.

The chatbot supports the rapid classification of faults. It bundles relevant machine data and document knowledge, presents correlations in an understandable way and suggests suitable next test steps. Decisions remain consciously with humans (human-in-the-loop). Machines are not controlled autonomously, except via explicitly approved workflows

Depending on the data source, the wizard can reference document locations/sections used and name the underlying measured values/periods. Role and authorization concepts, logging and clearly defined response formats also ensure traceability and trust.

Yes, depending on the requirements, a setup with local LLM is possible (e.g. for data protection/compliance). Alternatively, SaaS operation may make sense.

Through data minimization, clear data flows, authorizations and suitable operating models (on-prem/SaaS). The specific design is based on your architecture and your compliance requirements.