StoringsCheck & Kapuut: one knowledge base, multiple target groups

The context

Fault information in the installation industry is scattered across loose manufacturer PDFs and in the heads of experienced colleagues. This is slow, error-prone, and non-transferable: a new employee cannot interpret an error code independently, and knowledge is lost when an experienced technician leaves. In a safety-critical context (combustion, gas, flammable refrigerant), guessing is not an option. This tension came to light through a specific customer question from a national service and installation organization: can a customer service representative know immediately during a call what is wrong, how dangerous it is, and what the next step is?

Transforming the Dots has independently developed this concept as its own platform: a single knowledge base serving multiple target groups, from the agent on the phone and the technician in the field (later phase) to the consumer who Googles their error code.

The approach

The core of the platform is a verified knowledge base, built along a fixed hierarchy of category, brand, series, model, and error code. The key technical choices collectively determine reliability and scalability:

• Fixed source, not AI-invented. Each error code is taken verbatim from official manufacturer documentation, with source references per code from approximately 1,300 hosted documents. In a security-critical domain, this is the key distinction from a generic chatbot that can provide plausible but incorrect answers.
• AI as a layer above the source, never as a source. Approximately 69,000 AI-generated explanation blocks and a chat assistant (grounded RAG) make the knowledge base readable and searchable, with an automated quality check for style, attribution, and gas/CO consistency. AI unlocks, but is never the source of truth.
• Safety explicitly modeled. A controlled data field drives a traffic light model in the interface: safe, increased risk, or gas/CO hazard using a fixed protocol. The risk is visible at a glance and cannot be distilled from a block of text.
• Triage as a product, not just a lookup. A data-driven decision tree supports the entire customer conversation: what questions to ask, in what order, and when to escalate. That is where the time and quality gains lie for a customer contact center.
• One data source, multiple target groups. The same data serves the agent (StoringsCheck) and the consumer (Kapuut), with a visibility level per content block. Maintain one source, serve multiple markets, scalable without duplication.
• Multi-tenant white label, EU hosting. Each customer receives their own environment on a dedicated subdomain, with their own branding down to the design tokens and only their own service scope visible. Hosted on proprietary European infrastructure: GDPR-compliant and privacy-first as a selling point, not a cost.

Technical foundation: PostgreSQL with pgvector (relational data and semantic embeddings in a single database), Node.js, React, and Material 3 web components, running in Docker containers on a European VPS.

What it yielded

A working platform with two live portals on a single knowledge base: 47 brands, over 300 devices, and over 5,300 error codes (of which over 5,000 are verified) across six device categories, from central heating boilers and heat pumps to air conditioning and ventilation. StoringsCheck runs multi-tenant, with the first environments live and a pilot with the intended launching partner in preparation. Kapuut makes the same knowledge base accessible to the public and SEO-driven, with thousands of indexable error code pages for the consumer. More important than the numbers: the foundation is scalable. Adding a new category, a new brand, or a new customer involves entering data and configuring, not a redesign.