1. Background
The objective of the ITU AI for Good Sandbox Build-a-thon 5.0 is to harness crowdsourced inputs and actively involve vast communities, including students, researchers, and professionals, in building a collaborative and open knowledge base that enables cross-domain AI agent development. The use of AI agents is aimed at demonstrating the following:
- Application domains: enabling domain specific application use cases in 5G networks,
- Networks and services domains: Use cases in the network layers such as trouble shooting and resource optimizations
There is a growing need to understand how AI agentic solutions would function in various domains in the networks, services and applications. This involves both technology and strategies and innovations. This build-a-thon 5.0 challenges participants to build the foundational infrastructure — country-specific knowledge bases and AI agents — that will power realistic, policy-grounded applications.
This build-a-thon provides an opportunity for participants to collaborate globally in creating a shared AI knowledge infrastructure, contribute to a meaningful governance tool for the host country, and win exciting prizes.
2. Requirements for Participants
Participants are required to build and demonstrate an AI agent-based solution using a knowledge base created from use case related documentation and code in the domain of telecommunications, while adhering to host-country AI policies, standards and regulations.
Participants are invited to:
(1) Create innovative AI solutions in 5G, demonstrating the value of AI integration and agentic solutions in the application domains and networks and services domains,
(2) Make sure that the solution is mapped to and adhere to the published and open documents provided by the host country (including laws and regulations; national AI, digital and cybersecurity strategies; data protection and privacy legislation; institutional mandates and responsibilities; economic and industrial characteristics; labour-market conditions; intellectual property regimes; multilateral commitments and treaty obligations; energy and infrastructure constraints; and other relevant country-specific information)
(3) ITU reference code to extract, structure, and ingest data into a vectorized, distributed knowledge base hosted in the ITU AI for Good Sandbox.
(4) Use the cloud credits in the Sandbox to fine-tune AI models and create AI agents capable of one or more of the following roles:
- Innovative AI solution – an agent which uses the knowledge base and demonstrates unique value, solving problems in 5G.
- Policy and legal adviser — an agent that participants can query during a simulated error scenarios in any network layers (including applications, services), while adhering to applicable legislation, institutional responsibilities, emergency powers, international obligations, and sector-specific impacts.
- Institutional participant — an agent that assumes the role of a specific government actor (e.g., regulator, technical adviser, ministry) and contributes information or recommendations during the simulation.
- Multi-agent coordination — multiple agents representing different institutional roles (e.g., domain specific application agent / network and services agent in 5G, cross domain responsibilities such as health care, agriculture, education, innovations, security, telecommunications regulator, data-protection authority), each receiving different information, operating under different mandates, and producing independent assessments. Inter-agent coordination protocols should orchestrate the interaction.
(5) Design and administer interactive error scenarios with sequenced, chunked events released progressively (not all at once), reflecting realistic conditions where information is incomplete, unevenly distributed, and received at different times by different agents.
(6) Implement audit logging and chain-of-thought tracing so that agent decisions, reasoning, and policy references are transparent and reviewable.
(7) Participants are encouraged to publish their output including code repositories, knowledge base, gaps analysed as opensource. The open submissions would be considered to be made available in the ITU AI for Good Sandboxes.
The AI solutions developed in the Sandbox should be aligned with ITU-T Y.3172 and reflect the ITU AI Readiness Framework from (including but not limited to) the perspectives below:
- Open data and open-source management
- Generative AI content ecosystem
- Contextualization of AI solutions and regional development focus
- AI integration in domains and cross-domain analysis
- Human interaction and supervision in AI
- AI policy and implementation opportunity identification
- AI for social inclusion
- Digital infrastructure and energy considerations
3. Evaluation Criteria
Solutions will be evaluated on:
- Knowledge base quality — completeness, accuracy, and structure of ingested national material; ability to retrieve relevant policy context under query
- Agent capability — relevance and accuracy of agent responses during simulated error scenarios; ability to identify applicable solutions, while adhering to laws, mandates, and coordination mechanisms
- Policy, strategy, standards gap identification — quality of insights surfaced: unclear responsibilities, standards, technology and legislative gaps, coordination problems, conflicts between objectives. In addition to the country level policies, the relevance of the gaps and accuracy of the global policy examples would be a critical part of the evaluation. NOTE – Rather than analysing specific geographies or country specific recommendations, participants should focus on domain specific policy gaps and finding relevant global policy examples.
NOTE- The gaps analysed would include not only the host country but also the search space for gaps would include the expanded region with neighbouring countries. - Auditability — transparency of agent reasoning through logs, chain-of-thought traces, and source attribution
- Alignment with ITU standards — reflection of Y.3172 pipeline and AI Readiness Framework in the solution architecture
4. Benefits of Participation
- Access to online expert mentoring sessions
- Access to ITU AI for Good Sandbox — for selected build-a-thon participants
- Cross-domain collaboration with experts in industry, academia, and policymaking
- Hands-on experience building AI agents for real-world AI applications
- Deep dive into local AI integration priorities and practical contextualization of AI solutions
- Awards and certificates given by dignitaries from the host country
- Invites for AI for Good Global Summit 2027
- Opportunity for winning solutions to be deployed in the 2027 Summit live multi-country error exercise
- The participating countries will have access to the knowledge base as an output of the hackathon
- The participating countries will have access to those applications (including the agents) which are published as opensource, from the output of the hackathon.
- The gaps analysed would include not only the host country but also the search space for gaps would include the expanded region with neighbouring countries.
5. Awards and Certificates
The following awards would be given:
- First grand prize — 700 CHF
- Second prize — 450 CHF
- Third prize — 200 CHF
- Special prize for Women’s Team — 200 CHF
- Judges’ prize for teams — 450 CHF
The team should be a maximum of 4 people. For example: 3 participants + 1 internal mentor (optional). Only teams with a woman leader are eligible for the Special Prize for Women’s Team. All participants who submit valid solutions will receive a certificate of participation. A maximum of one prize will be awarded per team.
Awards would be distributed by the host country dignitaries.
Judges’ prize is awarded based on the level of engagement, future potential in terms of research, development, innovation, or standards potential, as determined by the judges’ panel.
6. Combined Training and Hackathon Structure
The hackathon will be combined with ITU AI for Good Sandbox training. The training provides the foundational knowledge and tools that participants apply during the hackathon.
Basic training (3 hours): Providing an entry-level understanding of AI and machine learning. The training centers around the ITU-standardized machine learning pipeline introduced in ITU-T Y.3172, focusing on understanding the vocabulary while getting hands-on experience with practical use cases. Essential information for each node in the pipeline is covered with concrete examples. Use case examples would be prepared as classwork in ITU format.
Advanced training (3 hours + 1 hour per domain): Providing domain-specific analysis for use cases based on the ITU-T Y.3172 machine learning pipeline and Y.3181 Sandbox. Participants work with the ITU AI for Good Sandbox environment to analyze domain-specific use cases from a standardized pipeline and agentic perspective. Key topics include the construction of country-specific knowledge bases, configuration of AI agents for policy analysis and error simulation, what-if scenario design, and policy improvement opportunities, standards identification and mapping. Standards and Policy opportunities will be identified, and AI strategy inputs will be derived as classwork for domain-specific use cases.
Agentic AI session (2 hours): A dedicated session on applying agentic AI to error simulation workflows, covering agent communication protocols, multi-agent orchestration for institutional role-playing, tool integration, safety guardrails, and validation of agent behavior in the Sandbox environment.
