Sustainable and futureproof development of rural areas in the Netherlands is one of the main challenges of the Dutch Ministry of Agriculture tackled with the Aanpak Ruimte voor Landbouw en Natuur (RLN). To achieve this, policy makers, provincial representatives, area managers and other stakeholders need to write and enforce policy based on reliable information. Unfortunately the vast amounts of information on the topic tend to make this process tiresome or even impossible.
As a solution for this problem, Gradient developed AIRE (AI for Regional Expertise) as a tool to enhance productivity and support policy enforcement at the Dutch Ministry of Agriculture. AIRE was created to provide reliable, data-driven answers to complex questions, helping policymakers, provinces, and advisory organizations make informed decisions. By leveraging AI-powered Retrieval-Augmented Generation (RAG), AIRE is able to extract valuable insights from a vast database of Dutch nature and agriculture-related documents maintained at Groen Kennisnet.
The project was completed within a span of four months, following an intensive co-creation process with early adopters and experts. Our goal was to ensure that AIRE delivers highly relevant, trustworthy responses while seamlessly integrating into existing workflows. The results of our pilot demonstrate the potential of AI-driven solutions in enhancing policy implementation and enforcement.
What is AIRE?
AIRE (AI for Regional Expertise) is an AI-driven Retrieval-Augmented Generation (RAG) application designed to enhance policy implementation and enforcement in Dutch nature and agriculture. Developed in a pilot project for the Dutch Ministerie of Agriculture, AIRE supports the enforcement of the Aanpak RLN (RLN) by providing accurate, contextual, and well-validated answers to complex questions. The system leverages open-source large language models (LLMs) and a database of approximately 500,000 documents to ensure reliable and relevant responses for policymakers, provinces, and advisory organizations.
Key Features of AIRE
- Domain-Specific Knowledge: Trained on a vast dataset of Dutch nature and agriculture-related documents to provide highly relevant insights.
- Retrieval-Augmented Generation: Combines retrieval-based search with generative AI to ensure responses are both factually grounded and contextually accurate.
- Expert Validation: Achieved an 8/10 grade based on validation by a group of five experts, ensuring high reliability.
- Collaborative Development: Built in close cooperation with early adopters, including policymakers and provincial representatives, to align the system with real-world needs.
- Scalability and Adaptability: Designed for easy expansion and integration into governmental digital workspaces and regulatory processes.
Our Development Journey
Developing AIRE was an iterative and collaborative process. From the outset, we worked closely with policymakers, provincial representatives, and advisory organizations to ensure that the application addressed real-world challenges. Using open-source LLMs, we built a robust RAG pipeline that efficiently processes and retrieves relevant information from a vast database of agricultural and environmental documents.
The pilot phase focused on validating AIRE’s capabilities in answering complex policy-related questions. Throughout the development process, we continuously refined the system based on expert feedback, optimizing accuracy, retrieval mechanisms, and response coherence. This co-creation approach allowed us to build an AI solution that is both technically advanced and practically useful.
What We Learned
The project provided several key insights:
- Expert Validation is Crucial: AI-generated responses must be continuously validated by domain experts to ensure accuracy and trustworthiness.
- Human-AI Collaboration Works Best: Policymakers and advisors benefit most when AI serves as an assistive tool rather than a replacement for expertise.
- Data Quality Matters: The reliability of AI-driven insights heavily depends on the quality and relevance of the source material.
- Customization is Key: Different stakeholders have varying needs, requiring adaptable and fine-tuned AI solutions.
Looking Ahead
Following the success of the pilot, we have agreed with the Ministerie of Agriculture to further develop and implement AIRE over the coming year in the Digitale Werkplaats. This next phase will focus on:
- Expanding the Knowledge Base: Incorporating more sources and refining document indexing for even greater accuracy.
- Enhancing User Interaction: Improving the interface to make AI-assisted policy research more intuitive.
- Integrating with Governmental Systems: Ensuring AIRE seamlessly fits into existing workflows and regulatory frameworks.
- Continuous Expert Validation: Establishing an ongoing feedback loop to maintain high response quality.
AIRE represents a step forward in AI-driven policy support, providing accurate and actionable insights for Dutch agriculture and nature management. Do you feel the same struggle of not being able to find the information that you know exists or if are simply interested in the AIRE tooling, feel free to contact us and schedule a cup of coffee.