Runko Works

Case study

From MVP to an enterprise knowledge-graph platform

Data from disparate legacy systems in one knowledge graph, so search, analytics and new applications plug into one platform instead of each source.

Client
Global energy technology company
Role
Lead software architect, team of up to 30
Period
2021 – 2025
Area
Platforms & data
Stack
Knowledge graphs · ontologies · microservices · cloud-native · ETL/ELT · Terraform · Kubernetes · GitLab CI/CD

Why it mattered

A large company keeps its data in systems that grew over many years: legacy databases, business tools, shared drives. Each of them grew on its own. There is no common standard, no way to tell that two records in two systems mean the same thing, and no common data model. Every new search, report or application has to integrate with each of them again, and every integration breaks a little differently.

A knowledge graph puts the data and the relationships between it in one place, so new consumers plug into one platform instead of each source separately. It is heavy artillery for data. It earns its weight where the links between the data, its meaning and its order decide whether a company sees the whole of what it knows.

The platform had proven itself as an MVP. The job was to turn it into something the business can run on, while the ground kept moving underneath: two large projects merged, infrastructure moved, and the company’s networks were rebuilt.

How it works

Data silos

  • Legacy systems
  • Business tools
  • Databases
  • Shared drives
  • No common standard
  • No identity of an entity across systems
  • No common data model

Ontologies of the data

Modelling
  • Data structure analysis
  • Ontologies
  • One model for every source

Processing along the ontologies

ETL / ELT
  • Process the data
  • Build the links
  • Build semantic objects

One place

Knowledge graphThe data and the relationships between it

Data registry

Explicit links between objects, kept for good

  • By rules
  • By hand
  • From outside

Interfaces

Consumers

  • APIOther toolsbuilt on the API
  • Search UI, like an online shopEnd usersdaily work in the UI
  • Export to the data warehouseAnalyticshundreds of dashboards
Heavy artillery for data. Links, meaning and order give the full picture of what a company knows.

From silos to one graph

  • Modelling. The data structures of the sources are analysed and described in ontologies: one model for structured and unstructured data, for tables and trees.
  • Processing. ETL and ELT pipelines extract and adapt the data from legacy systems, business tools and databases (MS SQL, Oracle, PostgreSQL, a cloud data warehouse), from shared drives, and directly from tools through their APIs and SDKs. Along the ontologies they build the links between entities and the semantic objects.
  • Loading. The result goes into one knowledge graph.
  • Data registry. Links between objects are also set explicitly: automatically by rules, by hand, or from outside. They become a permanent part of the graph.
  • Interfaces and who uses them. Other tools are built on the API. End users do their daily work in a search UI that works like an online shop. An export to a cloud data warehouse feeds analytics: hundreds of dashboards.

One platform instead of many consumers of many sources

  • A source-agnostic design: a search UI, an analytics layer and further applications consume the graph, not the sources. When the original programme changed course, the platform kept serving its other consumers.
  • Two large projects with several years of history each merged into one platform.

Infrastructure that can be operated

  • Parts of the system scattered across an OpenShift cluster and Windows virtual machines moved to the cloud as far as possible, under one architecture of cloud-native microservices.
  • An enterprise network carve-out: every network and connection rebuilt.
  • The project moved to infrastructure as code; DevSecOps and CI/CD practices established inside the existing teams.
  • Identity and access management for the platform inside a complex enterprise landscape.

People

  • Coordination of up to 30 people: a product owner, two external architects and two to three cross-functional development teams, with the architecture owned end to end.

Why this matters for AI

The same problems stand between most companies and useful AI in their business processes. An assistant or an agent is only as good as the data it can reach: if that data is scattered, inconsistent or locked in legacy systems, it will answer confidently from the wrong half. AI works on a reliable foundation of well-structured data, and building that foundation is the part I know how to do.