Pekka Kanerva
City of Lohja, a medium-sized city in Finland, has partnered with Miracle Finland to develop an advanced document chatbot that transforms how city employees access and use information stored in extensive internal document repositories.
“With the chatbot developed in the project, we will improve our organization’s internal communication and significantly facilitate and accelerate information retrieval for both supervisors and employees. Rapid access to up-to-date guidance streamlines daily work, harmonizes operating practices across units, and reduces the internal workload of support services (e.g. HR support, ICT helpdesk), thereby freeing up working time for other tasks and for the further development of operations.”
Pasi Perämäki
Director of Information Management and Development
City of Lohja
As the volume of unstructured data continued to grow, the City of Lohja required a solution that could move beyond traditional keyword-based search. In collaboration with Miracle Finland Oy, the city implemented an intelligent assistant running fully inside Oracle Database 26ai, ensuring security, performance, and data governance at the database level.
The user interface is built using the Oracle APEX low-code platform, providing a secure, responsive, and intuitive experience for employees. APEX enabled rapid development and deployment while eliminating the need for complex middle-tier infrastructure. Core application logic is implemented in PL/SQL, delivering high performance and maintaining data integrity directly within the database.
At the heart of the solution is a Graph RAG (Retrieval-Augmented Generation) architecture that significantly enhances information discovery compared to conventional search approaches. A Large Language Model (LLM) analyzes documents to identify key entities and their relationships, which are then stored and modeled using Oracle Property Graphs. This enables the chatbot to understand how municipal projects, decisions, and stakeholders are interconnected, rather than treating documents as isolated pieces of text.
In parallel, Oracle AI Vector Search is used to perform semantic similarity searches. By embedding document content as vectors, the database can retrieve relevant information based on the meaning and intent of a user’s query, even when exact keywords are not present.
By combining property graph traversal with vector-based semantic search, the chatbot delivers highly contextual and comprehensive answers. It not only identifies the most relevant documents but also uncovers related insights across the knowledge graph—capabilities that traditional RAG implementations often lack. The result is a powerful, database-native solution that saves time, improves decision-making, and enables city employees to fully leverage the collective knowledge stored in the City of Lohja’s data assets.