Misinformation Detection Tool

The starting point was a simple frustration. You see a post on Instagram or in a family group chat. Something about a local politician, a health announcement, an event that supposedly happened. It looks credible. You do not have time to verify it. You either share it or you do not, and either way you are guessing.

The goal was to remove the guessing. The concept brought to the session was a tool where you take a screenshot of any post and the tool tells you whether it holds up. Not an opinion, not another article. A verdict, with sources.

The first design decision was what to accept as input. A single Instagram post can have multiple photos, a caption, comments, overlaid text on the images. The team looked at a real Instagram post together and immediately ran into the problem. Which photo? The comments too? What if the claim is in the image text and not the caption? Trying to handle all of it at once would have killed the project before it started. The MVP decision was to accept one screenshot, one image, one claim at a time.

The second challenge was trust. A single fact-checking source has its own editorial blind spots. The tool was designed to cross-reference the Google Fact Check API, ClaimBuster, and sources like Nature and AP News in parallel, so the verdict is not coming from one place. The goal was to surface where sources agree and where they do not, rather than giving a single authoritative answer.

The third challenge was technical and still unresolved. Some of the Google APIs the tool depends on do not accept simple API keys. They require more complex authentication that adds infrastructure and time. The early sessions got the prototype working with mock data and a functioning upload-and-verify interface, but connecting the real APIs introduced security and configuration work that extended past the initial build timeline.

The prototype went from a spoken concept to a working interface in one session. The integration work is where the current challenge lives.