A workable AI hackathon idea starts with data that a team is allowed to use and a result that can be checked in a weekend. Government open data, published under a clear license such as the open government license of the Government of Canada (version 2.0), meets the first test, and public accessibility standards such as WCAG 2.2 give the second a measurable target. Ideas built on those two foundations are easier to finish, demonstrate, and defend to judges than ideas built on scraped or unlicensed material.
Key takeaways
- The Government of Canada's open government license permits copying, adapting, and commercial use of covered information, with an attribution statement.
- The license excludes personal information, third-party rights, and any suggestion of official endorsement.
- The AirNow API documentation states that its data are preliminary and not fully validated, so a project built on it must say so.
- WCAG 2.2 success criteria 1.1.1 and 1.4.3 give a hackathon team testable rules for text alternatives and color contrast.
- A strong demo shows a measured result on a small checked sample, not a claim that a model is accurate.
Check the data terms before writing any code
Most failed hackathon projects lose time on data, not on models. The Government of Canada's open government license is a worldwide, royalty-free, perpetual, non-exclusive license that allows information to be copied, modified, published, translated, adapted, and distributed for any lawful purpose, including commercial use. Users are asked to include the attribution statement the data provider specifies and, if none is given, a standard line saying the material contains information licensed under that license. The license does not cover personal information as defined in the Privacy Act, third-party rights the provider cannot license, official logos, or marks. It also bars uses that suggest official status or provider endorsement, so a demo should not borrow a department's name or crest. The information is provided as is, without warranty.
Three project ideas with their limits
The ideas below are illustrative. Each pairs an AI task with a source that has published terms and a limit that the team should state on the demo screen.
| Idea | Data or standard | Terms to respect | Main risk to disclose |
|---|---|---|---|
| Plain-language summaries of air quality readings | AirNow API: observations from more than 2,500 stations and forecasts for more than 500 cities | Data are preliminary and subject to change; official regulatory air quality data come from EPA's Air Quality System (AQS or AirData) | Summaries must not be used for regulatory or health decisions |
| Search assistant for open government dataset descriptions | Records published under the Government of Canada open license | Attribution statement; personal information excluded; no implied endorsement | A model may paraphrase a dataset description incorrectly |
| Accessibility checker for web pages | WCAG 2.2: 1.1.1 Non-text Content, 1.4.3 Contrast (Minimum) | W3C Recommendation, published in its December 12, 2024 edition | Machine-drafted alt text needs a person to approve it |
What the standards and APIs allow a team to measure
An accessibility checker is a good example of a project with a built-in scoreboard. Success criterion 1.1.1 requires that non-text content have a text alternative serving an equivalent purpose, with listed exceptions. Criterion 1.4.3 requires a contrast ratio of at least 4.5 to 1 for text, and at least 3 to 1 for large-scale text. A team can count how many images on a sample of pages lack alternatives, flag text that falls below those ratios, and have a model propose alt text for a human to accept or reject. Because the rules are numeric, judges can see pass and fail rather than a vague claim of improvement.
The AirNow documentation sets a different kind of boundary. It describes the service as intended for reporting and forecasting the Air Quality Index and says the data should not be used to formulate or support regulation, trends, or guidance. A summary tool built on it fits that purpose only if it presents readings as current conditions and links to official sources for anything else.
A scoping checklist for a 24-hour or 48-hour event
- Write one sentence naming the user, the task, and the data source.
- Record the license or terms for every dataset in the project README on the first day.
- Build a baseline without a model, such as a keyword search or fixed rule, so that the AI version has something to beat.
- Hand-label a small test set, even 30 items, and report how often the system matches the labels.
- Exclude personal information from inputs and from logs.
- Put the main limitation on the demo screen, not only in the README.
FAQ
Do hackathon projects need an open license on the data?
They need permission to use it for the purpose in question. Open government licenses state that permission plainly. Scraped pages and private datasets usually do not, and an entry built on data without rights is hard to defend to judges or later users.
Is it acceptable to build on an API whose data are preliminary?
Yes, when the demo says so. AirNow itself calls its data preliminary and subject to change, so a project that hides that fact misstates the source.
How should accuracy be reported at a hackathon?
As a count on a named sample, for example "correct on 24 of 30 hand-labeled items." A single percentage without the sample size tells judges little.
Can a project use the name of a government department?
Not in a way that suggests the department endorses it. The Government of Canada license prohibits uses that imply official status or endorsement.
Related reading
- AI event ideas for schools built on the AI4K12 Big Ideas
- Why lists of AI tools age quickly
- What Census data show about AI use by sector
- Counts and definitions for robots and AI in industry
More in AI Tools and Industry.
Last verified: October 2026
Sources
- Open Government License (Canada), version 2.0, Government of Canada
- AirNow API documentation
- Web Content Accessibility Guidelines (WCAG) 2.2, W3C Recommendation, December 12, 2024 edition