From policy to practice: How AI can strengthen public investment management in Ukraine
By Viktor Nestulia, Director of the DREAM Project Office, Open Contracting Partnership
Over the past three years, Ukraine has been implementing a comprehensive Public Investment Management (PIM) reform. That would be a significant undertaking in any country, Ukraine is doing it while hundreds of drones and missiles are launched at its cities every week. International estimates put the rebuilding costs at over US$ 588 billion. Thousands of projects that communities, regions, ministries, and dozens of international partners are working on simultaneously.
At the center of this reform is DREAM, Ukraine’s open digital ecosystem for public investment management. Built on open government and open contracting principles, DREAM connects state-owned and external registers, supports project monitoring and data analysis, and creates a single digital pipeline for public investment projects. DREAM is the digital backbone of that reform, supporting more than 12,900 public investment projects across 18 sectors and used by more than 1,300 municipalities, all regional administrations, all ministries and major state-owned enterprises.
DREAM covers the full investment cycle: from project initiation and feasibility appraisal through to prioritization, funding coordination and implementation monitoring. As the reform enters its full-scale implementation phase, we are exploring how AI can become its next practical enabler to strengthen institutional capacity, make complex PIM rules usable for every municipality in Ukraine, improve investment project quality, planning and delivery efficiency.
The scale of the challenge
That scale brings a practical capacity challenge. Public officials must navigate complex laws, methodologies, procurement rules, construction norms, financial requirements and donor conditions.
Local governments often operate with limited staff, fragmented expertise, and little time to navigate hundreds of pages of laws, methodologies, and manuals. For example, a full cycle of investment pre-feasibility study for a relatively small project takes about four weeks of full-time work for a single local government team. As a result, preparation of 10 projects becomes a huge challenge for a small local government.
During peak planning cycles, the DREAM Project Office is handling more than 850 support requests per day, even with extensive training, helpdesks and expert networks in place. In 2025 alone, we organized 96 training events covering almost 30,000 participants, helpdesks, and expert networks. Traditional support models struggle to scale at the pace this reform requires.
This capacity shortage is made acute by wartime pressures. Over 230 communities (around 15% of the country) have front-line status facing dense shelling, drone attacks, energy breakdowns and waves of internally displaced people (IDPs) on a daily basis. And all of them need to be included in the PIM process and get access to financing
Sound rules become hard to apply in practice when capacity is stretched this thin.
Our approach to AI
AI cannot replace public servants, expert judgment, or political accountability (at least yet). But it can dramatically improve how knowledge (policies, methodology, requirements, or sectoral expertise) is accessed, interpreted, and applied, especially in a fast-evolving, complex reform environment. That is the gap we are trying to close.
Within DREAM, we see AI not as an external automation tool, but as a support layer embedded directly into governance workflows. Its purpose is simple to explain but hard to achieve: to help users prepare and appraise better projects, understand requirements faster, and make more consistent, evidence-based decisions.
Three years of intense reform implementation have given us something most AI projects lack: a solid foundation. We have a verified knowledge base, a clear picture of where users get stuck, and the hard-won data to build an assistant that is useful. It’s not a fancy feature to brag about, but a real mechanism to support and empower those who have to rebuild our country.
With political support from the Ministry for Development and the Ministry of Economy, the DREAM Project office, backed by Open Contracting Partnership (with its expertise in providing AI Tools for public procurement) we joined forces with uStudio as IT and AI innovation partner, GCIEP and PwC (supported by FCDO) to develop an AI-powered assistant integrated into the DREAM ecosystem.
The approach is grounded in a working prototype, a curated and verified knowledge base, and a hybrid architecture aligned with Ukraine’s data sovereignty and ethical AI principles stated in White Paper on Artificial Intelligence Regulation in Ukraine.
DREAM AI Assistant functionality screenshot
The DREAM AI Assistant: Three stages of implementation
The rollout of the DREAM AI Assistant is strategically aligned with the PIM Reform Roadmap for 2024-2028, and advances through three distinct stages of implementation.
In 2026, the AI assistant will provide real-time guidance on PIM rules and the project preparation process directly within the DREAM interface. In this initial phase, we aim to bridge the gap between complex legal language and practical application by translating methodological requirements into accessible explanations tailored to each user’s specific role, sector, and location. Serving as a trusted, 24/7 entry point into the full PIM knowledge base, the DREAM AI Assistant provides accessible, practical guidance for government bodies at every level.
We tested the DREAM AI Assistant in April 2026. Focus group participants representing communities have used the assistant for the project preparation process, and provided a medium score of 4.33 out of 5 points. While we continue to tailor and upgrade the assistant, it has already become obvious that it will reduce workload and speed up the procedures guiding DREAM users through the project preparation process.
In 2027, the DREAM AI assistant will expand into quality assurance, helping appraisal teams identify missing indicators or inconsistencies in feasibility studies and suggesting ways to better align projects with national sectoral priorities, improve their economic, environmental, financial and other cases. By comparing new submissions with historical validated cases, the system will enhance the consistency and transparency of thousands of investment decisions.
In 2028-2029, the aim is advanced investment modeling using geographic, demographic, fiscal and other data to help officials explore various development scenarios and model alternative investment options. This functionality will allow for scenario-based assessments to determine where public resources can achieve the greatest possible impact.
Building a safe and responsible AI architecture
Because public investment decisions involve high stakes and sensitive data, the DREAM AI assistant is built on a hybrid and fully auditable architecture designed to ensure responsible governance.
The core of the system utilizes Retrieval-augmented generation (RAG), grounding every response exclusively in verified laws, manuals and guidelines. To safeguard sensitive information, the system employs local language models hosted on secure infrastructure to protect data sovereignty. Every answer is traceable to an approved source, timestamped and version-controlled. Furthermore, the architecture includes a human-in-the-loop validation process to ensure that all content remains accurate and reflects the most recent regulatory updates.
To minimize the risk of AI hallucinations or the use of outdated guidance, every response generated by the system is traceable to an approved source and marked with a specific timestamp and version. Potential risks, such as factual errors or privacy concerns, are proactively addressed through structured prompt design and rigorous access controls.
A dedicated multidisciplinary oversight team monitors the system to ensure it adheres to ethical standards and remains focused on its public purpose.
Ukraine’s lesson for the world
Ukraine’s experience shows that effective AI in government does not start with algorithms. It starts with clear rules, strong institutions, and the discipline to design technology that serves public purpose. The AI layer being added to DREAM is only as useful as the reform infrastructure beneath it, and that infrastructure took three years and enormous effort to build.
If done right, this initiative can become not only a catalyst for Ukraine’s recovery and development but a globally relevant test case for how AI can responsibly strengthen public investment management systems in complex environments. A country rebuilding under fire, with digital governance already at the cutting edge, is exactly the right place to prove what is possible.
