Global AI Builder’s Challenge

About the challenge

Your idea should do more than sound impressive. Build something people can actually use.

Global AI Builder’s Challenge is a free online hackathon hosted by Refactor The World for students aged 13 and older. From November 21–28, 2026, work independently or in a team of up to six to create a new project or meaningfully improve an existing one.

Explore what AI can make possible: a tool that helps someone learn, an assistant that handles a frustrating task, a more accessible experience, a creative application, a research tool, a game, or another project with a clear purpose.

AI integration is strongly encouraged and is reflected in the judging rubric. Thoughtful use of AI during development also earns credit. AI is not an entry requirement, although projects without AI will need exceptional performance elsewhere to remain competitive.

We care about a project’s practical value, its use of AI, whether it works, and how clearly people can use it. You do not need a large team, expensive tools, or a complicated idea to build something worth recognizing.

We will recognize the Top 10 overall projects and winners across additional award categories. Winner certificates are confirmed. Additional prizes may be announced if sponsorship arrangements are finalized.

Participation is free. Submit a public GitHub repository, a 2–4 minute demonstration video, and a Devpost project page identifying every team member and explaining what you built or improved.

The event starts November 21, 2026, at 12:00 a.m. PST. The submission deadline is November 28, 2026, at 3:00 p.m. PST (America/Los_Angeles). Winners are scheduled to be announced on December 5, 2026, at 3:00 p.m. PST, within one week after submissions close.

Judging method

Use four criteria, each scored from 0 to 5. Each criterion contains five checks. Each satisfied check earns one point. An unmet or unsupported check earns zero points. Do not use half-points for an individual judge’s checklist.

Judges are assigned to specific criteria. More than one judge may score the same criterion. Scores from those judges are averaged within that criterion before its weight is applied. Every evaluated project needs a score for all four criteria. A missing evaluation is not a zero; obtain the missing evaluation.

Weights: Problem Value & Practicality 20%; AI Integration & Development 30%; Functionality & Reliability 30%; Design & Usability 20%. Let V, A, R, and D be the averaged scores, each between 0 and 5. Final score out of 100: 4V + 6A + 6R + 4D. A criterion’s weight does not increase because it has more judges.

V evaluates the proposed benefit and feasibility of adoption. A evaluates the use of AI in development and the implementation of AI within the project. R evaluates actual behavior and reproducibility. D evaluates how clearly a person can understand and operate the experience.

The same evidence may inform different checks, but points are not awarded or deducted twice for the same underlying quality. Video production quality, presentation confidence, team size, and spending are not separate judging criteria.

Checklist scores are recorded externally using Devpost’s offline judging option. This refers only to the scoring method; the hackathon remains online.

Problem Value & Practicality — 20%

Evaluate the proposed value for the intended user, assuming the described capabilities work. Score implementation behavior under R and interaction quality under D.

Award one point for each satisfied check:

• V1 — Specific user: The project identifies a particular intended user and a concrete situation in which they would use it.
• V2 — Supported need: A credible observation, firsthand example, user comment, or cited source supports the stated need or desired experience.
• V3 — Solution fit: The proposed core capability directly addresses that need or experience.
• V4 — Meaningful benefit: The project explains a concrete advantage over a reasonable current alternative, workaround, or lack of access.
• V5 — Practical adoption: The intended users could realistically access the necessary equipment, connectivity, data, permissions, and operating resources.

A small audience can earn full credit. Entertainment, creativity, and personal usefulness count as legitimate benefits. A large claimed market does not automatically earn more points.

Score: 0–5.

AI Integration & Development — 30%

Evaluate the team’s use of AI and the implementation choices connecting AI to the project. Score the project’s practical benefit under V, actual results under R, and interaction quality under D.

Award one point for each satisfied check:

• A1 — AI-assisted development: The team provides a concrete example of AI contributing to work completed during the event and explains how a team member reviewed, tested, or adapted that contribution.
• A2 — Defined AI role: The finished project implements an identifiable AI task, and the team explains why an AI approach is appropriate for that task.
• A3 — Input preparation: The implementation deliberately selects or prepares the inputs, context, retrieved information, or data supplied to the AI.
• A4 — Task-specific configuration: The implementation includes prompts, instructions, model configuration, training, or other controls tailored to its intended AI task.
• A5 — Output integration: The implementation contains a defined mechanism that passes AI results into a project feature, workflow, or action.

A1 can be earned through AI-assisted development. A2–A5 require AI integrated into the finished project. For A2–A5, assess the implementation and its purpose. Assess whether it produces correct, repeatable results under R. Using a particular provider, paying for a model, training a new model, or fine-tuning is not required.

Score: 0–5.

Functionality & Reliability — 30%

Evaluate what the submitted project actually does. Score the value of its purpose under V, its AI implementation choices under A, and clarity of interaction under D.

Award one point for each satisfied check:

• R1 — Complete main workflow: The project can carry out its main task from start to finish and produce a usable result.
• R2 — Result quality: Representative outputs meet the project’s stated requirements, supported by expected results, reference examples, or another suitable evaluation method.
• R3 — Repeatability: The main workflow succeeds across repeated runs or representative cases. Where outputs vary, they still meet the required behavior.
• R4 — Failure handling: At least one relevant invalid input, dependency failure, or foreseeable failure condition receives a controlled response rather than false success or an uncontrolled crash.
• R5 — Reproducibility: The submitted code, dependency information, configuration guidance, and instructions allow another reviewer to reproduce the demonstrated behavior, with external service requirements identified.

Evaluate the submitted version. Clearly disclosed limitations should affect the relevant checks rather than automatically removing unrelated points.

Score: 0–5.

Design & Usability — 20%

Evaluate how clearly a person can understand and operate the project. Score whether the underlying functions execute successfully under R.

Award one point for each satisfied check:

• D1 — Clear entry point: The interface or immediate instructions make the project’s purpose and starting action understandable.
• D2 — Clear task flow: The steps for the main task follow a clear sequence without unnecessary navigation or repeated input.
• D3 — Useful feedback: The experience communicates relevant status, progress, and completion in understandable terms.
• D4 — Readable presentation: Text, controls, and results are legible, distinguishable, and presented in a format appropriate to the project.
• D5 — User control: Users can change inputs, restart, cancel, or repeat the main task through appropriate controls or instructions without developer intervention.

Apply these checks appropriately to websites, apps, command-line tools, games, and hardware experiences. Elaborate visual styling is not required. D5 evaluates whether an understandable control path exists. R evaluates whether the operation behind it works.

Score: 0–5.

Awards overview

We will recognize the Top 10 overall projects and winners across 17 additional award categories.

Winning team members will receive certificates recognizing their awards. Certificates are reserved for winners.

A project may receive one overall placement and multiple category awards. Category awards do not change a project’s overall score or ranking.

Additional prizes are being discussed with potential sponsors. Cash, credits, merchandise, and other sponsor benefits are not currently guaranteed. Any confirmed additions will be announced on the hackathon page.

The first 15 category awards (Best AI Agent through Best AI Engineering) require AI integrated into the submitted project. AI-assisted development alone does not qualify a project for those awards. Best Design & User Experience and the Originality Award do not require AI integration.

A project can fit multiple award categories. These categories describe different forms of recognition; they are not additional overall scoring criteria.

Sponsors / Promotional partners

Sponsorship discussions are ongoing. Confirmed sponsors and any additional prizes or benefits will be announced here when arrangements are finalized.

Organizations interested in supporting Global AI Builder’s Challenge can contact refactortheworld@gmail.com.