At the end of June, we successfully completed the first milestone of the midPilot project, funded by the European Union’s Recovery and Resilience Plan. This milestone, running from April to June 2025, focused primarily on research and analysis of the current state, combined with a series of experiments, measurements, and prototype developments. These activities laid the groundwork for the first version of the project’s architecture and design. A detailed summary of our findings and outcomes is available in the published milestone report and concluded in the following sections.
Research, Analysis, and Prototypes Scope
The main focus was to explore how AI can enhance midPoint’s capabilities, particularly in the areas of connector code generation, mapping recommendations, and correlation logic. We started with research in state-of-the-art AI technologies, including Large Language Models (LLMs) and other AI/ML techniques, to identify their potential applications in midPoint. Since the midPilot project scope is quite broad, the research and analysis included areas for scraping, data extraction, and data transformation, as well as the use of AI in code generation and recommendations for configuration. Based on research findings, we conducted a series of experiments and measurements to evaluate the feasibility and effectiveness of various AI techniques in these areas. We also developed several prototypes to validate our ideas and approaches, including a connector code generator, a model-mapping recommendation system, and a correlation recommendation system. The results of these experiments and prototypes are documented in the milestone report, which provides a comprehensive overview of our findings and outcomes.
In addition to prototyping and exploring state-of-the-art AI technologies, we also dedicated time to analyzing and designing changes for midPoint, midPoint Studio, and the Connector Framework (ConnId) itself. We focused not only on the integration capabilities for AI but also on the overall usability and user experience of midPoint, including the design of web-based wizards and an IntelliJ IDEA Studio plugin. These results are also described in the milestone report.
Our next steps are to validate the results from Milestone 1 and prepare a minimum viable product (MVP) to test how all the components work together in practice.
Challenges – and How You Can Help
During our experiments, we encountered a challenge related to limited access to real-world configuration data. While we have some publicly available configurations from our mailing list and support portal, more diverse and representative data would make our experiments and algorithms more effective. If you are willing to share your configurations, please send them to aidata@evolveum.com. We do not require sensitive or production data. It would be an ideal situation if we had such real-world data from the real environment, but we understand that in the space of IGA it means sharing sensitive data. Therefore, we are asking for configurations that do not contain sensitive information. The approach we’ll take is to generate synthetic data based on the configurations you provide. The most valuable inputs are:
- Resource configurations: mainly the schemaHandling part, including configuration for mappings, synchronization, correlation, and associations.
- Optional: roles, policies, objectTemplates, or at least the mappings you use in your environment.
This data will not be used for AI training, but solely for analysis, to better understand typical real-world setups. If you have any questions or concerns about sharing your configurations, please feel free to reach out to us at aidata@evolveum.com.
Why Your Contribution Matters
Having access to more configuration examples would allow us to:
- Identify common patterns in scripts and mappings, and integrate them directly into midPoint.
- Replace repetitive Groovy scripts with built-in functions for declarative configuration.
- Adjust default behaviors based on frequently used synchronization settings, speeding up onboarding.
- Detect complex, recurring solutions that could inspire new midPoint features.
- Develop heuristic algorithms to guide users in navigating and configuring their environments more effectively.
All with one aim – to streamline the integration process and improve the overall user experience in midPoint. Therefore, your contributions could directly influence the usability, efficiency, and capabilities of future midPoint versions.
Publicly Available Resources
One of the key outcomes of this milestone is the open source no-code/low-code connector framework, designed to create connectors for cloud applications offering either a REST API or SCIM 2 interface. The framework is already available in our GitHub repository. In its current version, it supports basic operations such as reading objects from an application. Additional capabilities – including creating, updating, and deleting objects – will follow in future iterations.
This framework will serve as one option for building new connectors in midPoint, potentially with AI assistance. However, it can also be used without AI, providing a long-term solution for developing custom connectors in both current and future midPoint versions.
Conclusion and Next Steps
Milestone 1 has given us a strong foundation. With the new connector framework in place and comprehensive research conducted, we are moving towards Milestone 2. Internally, we decided to divide this milestone into two phases, starting with the MVP to bring all the pieces together.
This project has received funding from the European Union through the Recovery and Resilience Plan of the Slovak Republic.
