Veripulse

Inspiration

During emergencies and unexpected incidents, people often turn to social media or community platforms to understand what is happening around them. The problem is that crowdsourced reports can be incomplete, duplicated, misleading, or difficult to verify. With AI-generated images and content becoming increasingly realistic, simply seeing a photo attached to a report is no longer enough to establish trust.

This inspired us to build VeriPulse — a platform that helps turn crowdsourced incident reports into more useful, trust-scored information without treating AI as an absolute judge of whether something is true or false.

What it does

VeriPulse allows users to submit incidents happening around them, including a description, location, category, and supporting evidence such as an image.

Instead of simply displaying every report equally, VeriPulse analyzes multiple signals associated with a submission and generates a trust score. These signals can include:

Completeness and consistency of the incident description
Supporting evidence attached by the reporter
Image metadata when available
Location and time consistency
Similar or duplicate reports from other users
AI-assisted analysis of the submitted information

Reports can then be displayed on a dashboard so users can quickly understand what is happening and identify incidents that may require additional verification or attention.

Importantly, the trust score is not intended to declare that an incident is definitively "true" or "false." It provides additional context based on available evidence.

How we built it

We designed VeriPulse as a full-stack incident-reporting platform with an AI-assisted verification pipeline.

When a user submits a report, the application collects the incident information and any supporting evidence. The backend processes these inputs and extracts relevant verification signals.

We integrated Gemini to help analyze unstructured incident descriptions, identify useful information, categorize reports, and assist with evaluating the completeness and consistency of submissions.

For uploaded images, our verification approach considers available metadata and other contextual signals rather than assuming that an image alone proves an incident occurred.

The results are combined into a trust-scoring system and presented through a dashboard that makes reports easier to understand and prioritize.

We also used GitHub for version control and team collaboration throughout the hackathon.

Challenges we faced

Our biggest challenge was answering a fundamental question: How can a system determine whether a crowdsourced report is trustworthy?

We quickly realized that relying entirely on AI would create another problem. AI can make mistakes, and modern generative AI can create convincing fake images and descriptions.

We therefore shifted our approach from trying to "prove" whether a report is real to evaluating multiple independent signals and presenting a confidence-based trust score.

Another challenge was image verification. Metadata such as EXIF information can provide useful clues about when or where an image was captured, but metadata can also be missing, removed, or modified. Because of this, we designed metadata to be one verification signal rather than definitive proof.

Building and integrating these components within the limited time of a hackathon was also a major challenge.

What we learned

Building VeriPulse taught us that responsible AI systems should communicate uncertainty instead of presenting AI-generated conclusions as absolute facts.

We learned how generative AI APIs can be integrated into a real-world application, how different verification signals can complement each other, and why privacy and responsible data handling are especially important when working with location data and user-generated evidence.

Most importantly, we learned that AI can be more valuable as a tool that helps people evaluate information rather than as a system that simply decides what is true.

What's next for VeriPulse

With more time, we would expand VeriPulse with cross-report verification, stronger duplicate detection, improved image authenticity analysis, real-time geographic incident visualization, reporter reputation signals, and better methods for detecting manipulated or AI-generated media.

Our long-term goal is to make VeriPulse a transparent and privacy-conscious platform that helps communities make sense of crowdsourced information while keeping humans in control of the final judgment.