+15 municipalities
The Cirrus platform supported circular economy decision-making across, from Helsinki to Istanbul.
EU Horizon
The platform's credibility and impact were recognized through funding under the European Union.
The Challenge
Building a platform that serves diverse municipalities across Europe — each with their own data sources, languages, and circular initiatives — presented a unique set of engineering and data challenges. The core problem was not a lack of information, but the inability to access, connect, and trust it at scale. Three challenges stood in the way:
- Data fragmentation across languages and formats: Circular economy strategies, regulations, and local initiatives were scattered across unstructured documents, different languages, and isolated silos. Without a unified taxonomy or shared framework, cross-border learning and structured data access were effectively impossible.
- Hallucination risk in a high-stakes context: For public policy and urban planning, AI accuracy is non-negotiable. Standard large language model approaches carried too much risk of generating misleading or ungrounded outputs — a critical concern when city planners rely on the system to make consequential decisions.
- Scalability for non-technical users: The platform needed to digest and reason across massive, heterogeneous datasets while remaining intuitive and accessible to policymakers and city planners with no technical background.
Our Approach
Following a design sprint to map the data landscape and define the platform architecture, Enjins and Metabolic co-developed Cirrus — an AI-enabled intelligence engine that transforms scattered municipal documents into structured, queryable knowledge.
- Semantic Data Structure and Unified Taxonomy: Rather than feeding raw data into a model, we digitized Metabolic’s proven systems-thinking framework into a standardized taxonomy. This gave diverse city datasets a common language, enabling meaningful cross-border comparisons and structured knowledge retrieval for the first time.
- Production-Grade Retrieval Augmented Generation: We implemented an advanced RAG system that goes far beyond basic keyword search. By using vector embeddings to capture the semantic meaning of a user’s query, the platform retrieves the most relevant local and global context to answer complex policy questions — combining global CE knowledge with city-specific data for genuinely tailored responses.
- Multi-Layered AI Safety and Trust: To ensure the platform met the reliability standards required for government use, we engineered a comprehensive safety protocol. This included strict prompt engineering to frame interactions responsibly, domain-specific guardrails to keep responses focused on circular economy topics, and continuous monitoring to detect and address issues in real time — ensuring the AI acts as a reliable expert rather than a generative tool left unchecked.
- Continuous Improvement Through User Feedback: The system was designed to evolve. By incorporating user feedback and new data on an ongoing basis, the platform adapts to the changing landscape of circular economy knowledge and the unique needs of each city it serves.
"With this project, we are consolidating all relevant circular economy knowledge into one dedicated platform."
Jan van 't Hek
Project Manager at Metabolic
