The team behind Kelaria — RAUSCH Technology
AI in production at Vodafone · AI projects with the German Energy Agency (dena) · hosted in the EU region Frankfurt · publicly funded research
Kelaria makes your company's domain and experience knowledge reliably usable: evidenced, traceable and embedded in your products and processes. What sits in people's heads becomes an asset that stays and that you can sell.
Hardware can be copied. Knowledge can't.
Three developments are hitting mechanical and plant engineering at the same time. All three lead to the same asset: the knowledge about your own machines.
Your most experienced service technician can hear a fault in any machine. In three years he retires — and his knowledge is documented nowhere. Who will still know why a particular machine always fails in summer?
Within 18 months a competitor rebuilds the hardware and the price advantage is gone. What remains is the knowledge about your own machines — provided it was made tangible.
Your customers expect more than the machine today: availability, remote support, data-driven services. Whoever offers no digital business or service model loses the customer interface to whoever does. The foundation for it is usable, reliable knowledge about your own equipment.
When hardware becomes a commodity, service is the business. A traceable knowledge system turns the service cost factor into new, sellable models.
Fast, reliable diagnosis makes operator and availability models possible — uptime guarantees, pay-per-use. Recurring revenue instead of one-off sales.
The knowledge assistant, embedded in your customer portal: end customers solve issues themselves — monetizable as a premium feature.
New markets, languages, locations — service grows with the knowledge graph, not with the number of your senior technicians.
Every answer with source and reasoning. Only that makes AI usable in liability- and safety-critical service — and defensible in front of the board.
No chatbot, no one-off project, but an expert knowledge system: it combines structured domain knowledge, your experts' rules and search across your documents into an answer with source and reasoning. More than search, and it learns with every document.
Manuals, documentation, service knowledge — via upload and API today, further source connectors in progress. Every new document is processed immediately.
Specialized agents extract entities, relationships and rules — sharpening the knowledge graph and RAG with every pass.
The agents evaluate every new document and present the results to your domain experts for review. Human-in-the-loop: what counts gets confirmed.
Kelaria delivers answers with source and reasoning — embedded in your systems, not as a standalone app.
Kelaria builds the knowledge, maintains it and activates it — right where the work happens. A little better with every document.
Behind the evidenced answers runs a curated knowledge graph, the engine room of the system. Here your specialists build, review and query it. A look at a real graph from a live project.
From the error code via the affected component to the possible causes — the chain a technician holds in their head, as a visible path in the graph.
Question in, answer out — with numbered citations and a reasoning trace. Every statement leads back to its source.
IF→THEN against the real values in the graph, with a live match count. Your experts write, Kelaria evaluates.
The next expansion step we are building right now: Kelaria searches the graph it has grown while no one is watching — and brings the findings to your experts each morning: confirm, adjust, discard.
“These symptoms always occur together” — Kelaria spots the cluster and turns it into a rule proposal.
Two components that ought to belong together — but have no edge yet. Kelaria proposes it.
Outliers in the graph — orphaned nodes, unusual combinations — that deserve a second look.
Once live data is connected: measurement series that drift out of rhythm — fed back into the graph as a symptom.
The processing pipeline already runs today with every new document — the autonomous pattern and anomaly discovery is the expansion step we are on right now. Together with your experts, never on its own.
Kelaria connects to the systems your knowledge already sits in — via open interfaces and MCP, with no data migration. Whatever is new, the pipeline keeps pulling in.
Available today: document upload and open APIs. Further connectors — SharePoint, Confluence, Jira, ERP, web crawler — are in progress; the order follows your stack. Additional integration via MCP.
“The AI delivers powerful heating-monitoring capabilities — end customers reduce energy consumption and CO₂ sustainably.”
— Dr. Sebastian Groß, VodafoneThe same mechanism, a different knowledge base — what we learned about heating systems, we're now building for machines and plants.
A living knowledge system pays off not in theory, but at four points in mechanical and plant engineering — which we work out concretely together in the AI-Case-Canvas.
The right answer on the first visit — instead of a second trip.
Diagnosis in minutes instead of hours — the knowledge is instantly available.
The skill of experienced staff is secured before they retire.
New service staff are productive in weeks, not years.
What the mechanism can deliver is documented: at Vodafone, 24% energy savings for the end customer. We calculate the concrete ROI in the AI-Case-Canvas — with your numbers, honestly.
We didn't build this for one industry, but as a mechanism. What we proved in heating transfers — the architecture stays, only the knowledge base is swapped.
Machine data is corporate capital. Kelaria runs today in the EU region Frankfurt — GDPR-compliant, with a data processing agreement and EU standard contractual clauses; LLM inference stays in the EU. And because the knowledge lives in the graph, not the model, your knowledge graph belongs to you — exportable, not locked into a model. The fully sovereign tier runs on-premises on a German open-source stack.
Citations and a reasoning trace on every answer. No black-box chat without evidence.
Your experts encode IF→THEN rules. Kelaria evaluates them deterministically against the real values in the graph and runs traceable diagnosis chains — no probability guessing, but evidenced conclusion.
AI agents evaluate every new document, domain experts confirm — the knowledge grows without any project standstill.
Kelaria docks onto existing AI systems via MCP — ChatGPT, Claude, Langdock & co. No rip-and-replace, no lock-in.
Managed in the EU region Frankfurt, customer cloud in your account, on-premises on German open source — all three available. Sovereignty is architecture, not a surcharge.
The knowledge foundation behind Kelaria emerges in publicly funded research projects — peer-reviewed, not asserted.
14 months live at Vodafone. We show what's running — not what could run.
“We build AI that actually runs in the industrial mid-market — not on a slide. Kelaria is our expert knowledge system for industry: the same mechanism we proved in heating, for your machines.”
Engineers and domain people under one roof in Würzburg — the hands that build the knowledge system behind it. No stock photos.
In 20 minutes we sketch your concrete use case and an honest effort estimate, together. We'll even answer the “build it yourself or with us” question straight, even when the answer is “build it yourself.” You invest time, nothing else.
No sales funnel. You receive a use-case sketch you can take forward internally.