Postcard

Honesty-as-a-service

Comment 2

Postcard: Devpost submission

Tagline: Democratizing the truth. Trace every post back to its source. GitHub: github.com/postcardhq/postcard Demo: postcard.fartlabs.org

Inspiration

In our current "post-truth" era, information is often consumed via contextless screenshots. By the time a post goes viral, it has been cropped, captioned, and stripped of its metadata—making it nearly impossible to tell if the content is authentic or a manufactured narrative. We wanted to build a tool that acts as a digital "way back machine" for credibility, restoring the lost link between viral content and its primary source.

What it does

Postcard is a digital forensics pipeline designed to rebuild trust in online media. It takes a social media URL (or a screenshot) and traces it back to its original source. The core of the project is the Postcard Score (0–100%), a credibility metric calculated by auditing how much the content has drifted from the ground truth.

Key Features:

  • Forensic Traceability: Automatically identifies the primary source of a claim across X, Reddit, and major news outlets.
  • Drift Analysis: Audits content for forensic consistency, checking for temporal alignment and attribution errors.
  • Screenshot-to-URL Resolution: Part of our original vision, this feature allows users to upload a screenshot to find its live, interactive counterpart for deep verification.
  • Sub score Breakdown: Users don't just get a number; they see a breakdown of source reliability, temporal verification, and cross-platform corroboration.

How we built it

We developed a 4-stage forensic pipeline focused on deep audit log generation:

  1. Multimodal Ingest: We utilized Jina Reader to ingest live content and establish a "ground truth" version of the post.
  2. Forensic Audit: Using Playwright, we performed direct site checks to verify the origin and ensure the timestamp aligns with the reported narrative.
  3. Corroboration Engine: We implemented a deep search across trusted domains to verify claims and find mentions of the content elsewhere to determine its "drift."
  4. Verification Platform: Built with Next.js and Tailwind CSS, providing a clean, terminal-inspired interface for quick, simple forensic verification.

Challenges we ran into

The biggest hurdle was the "Wisdom of the Crowd"—trying to find a live URL from a static, often low-quality screenshot by triangulating platform-specific clues. Social media platforms also have aggressive bot detection, which made using Playwright for forensic verification difficult. We had to fine-tune our headers and navigation patterns to ensure we could retrieve the necessary metadata without being blocked.

Accomplishments that we're proud of

We successfully built a functional 4-stage pipeline that does more than just "fact-check"—it performs a technical audit. Seeing the Postcard Score update in real-time as the corroboration engine found matching sources across different platforms was a huge "aha!" moment for the team.

What we learned

We gained a deep appreciation for the fragility of digital metadata. Information is easily manipulated once it leaves its native platform, and screenshots are the primary vector for "context stripping."

Key insights:

  • LLMs as Forensic Auditors: We learned that multimodal LLMs can be used for more than generation—they are powerful forensic tools for drift analysis (comparing "ground truth" to "viral claims").
  • oEmbed as a Truth Source: Traditional scraping is often blocked or low-fidelity; leveraging official oEmbed APIs is a more robust way to capture absolute metadata (exact timestamps and author IDs) for OSINT verification.
  • Wisdom of the Crowd: Rebuilding a credible forensic trail from a static, low-quality screenshot is difficult, but multimodal agents can bridge the gap by triangulating platform-specific clues found across the digital landscape.

Key Metrics

  • Forensic Pipeline Stages: 4
  • Postcard Score Range: 0-100%
  • Type Safety: 100% (TS/Lint)
  • Supported Platforms: X, Reddit, News Articles, Screenshots

Judging Alignment

  • Cybersecurity: Demonstrates digital forensics and OSINT techniques to verify data integrity.
  • Innovation: Uses multimodal LLMs for "drift analysis" rather than simple text fact-checking.
  • User Experience: Provides a clean, terminal-inspired interface with real-time polling updates (migrated from SSE to ensure 100% production stability).

What's next for Postcard

  • Browser Extension: Bringing Postcard directly into the feed, allowing users to right-click any post to see its score instantly.
  • Expanded Platform Support: Increasing our corroboration engine to include specialized platforms like Telegram and Discord.
  • API for Journalists: Opening up our forensic pipeline as an API for newsrooms to verify user-generated content in real-time.

Demos

Built With

Share this project:

Updates