Logo Clover Dynamics companyUnlock AI Potential
Logo of Clover Dynamics
Logo of CookRoom by Glovo

AI-Powered Browser Extensions for Real-Time Evidence-Backed Verification of Political Videos

FactFlow develops real-time verification technology for political speech, placing sourced verdicts beside the video a person is already watching across a cross-browser extension and public web/PWA experience.

Country:
USA
Technologies:
TypeScript
Next.js
Whether you're modernizing existing operations or building something entirely new, Clover Dynamics gives you the AI-native engineering support needed to move faster, automate deeper and stay ahead
Want the complete technical story? Download the full case study.
AI-Powered Browser Extensions for Real-Time Evidence-Backed Verification of Political Videos
The Challenge

Political claims spread at the speed of live media, while conventional verification usually arrives after the viewing moment. FactFlow needed verification to appear directly on YouTube and X without asking users to leave the video or trust an opaque AI answer. The system had to work with captions, inaccessible tracks and live audio; preserve original speech and evidence; survive Manifest V3 service-worker eviction; adapt to volatile host-platform DOMs; control real-time processing costs; and protect tokens, captured audio, transcripts, telemetry and third-party credentials across trust boundaries.


background image
Accelerate ahead of competitors with machine customer capabilities, agentic commerce and AI-native systems built from the ground up for autonomy and scale
The Solution

Clover Dynamics created an evidence layer that follows the viewer. The extension detects supported video, chooses between previously processed content, caption-first processing and live tab-audio fallback, submits timestamped claims to the existing verification platform, and renders verdicts beside the video with explanation, evidence and citations.


Deepgram streaming speech-to-text, deterministic phrase assembly, Silero VAD and an optional bounded Gemini 2.5 Flash boundary gate accelerate claim extraction without making AI the authority. Service-worker, offscreen-document, extension-page and content-script responsibilities are aligned with Manifest V3 lifecycles. Authenticated Socket.IO and LiveKit delivery persist rooms, validate payloads, deduplicate verdicts, buffer temporary disconnects and recover after reconnect, while silence gating, caps, covered-range skipping and privacy-aware telemetry control cost and risk.

The Solution
Case study diagram

The Features

In-Video Verdict Experience
In-Video Verdict Experience
Live Verdict Rail and Ticker
Live Verdict Rail and Ticker
Resilient Manifest V3 Runtime
Resilient Manifest V3 Runtime
Evidence-Backed Detail Panel
Evidence-Backed Detail Panel
YouTube, X, and Cross-Browser Support
YouTube, X, and Cross-Browser Support
Privacy, Cost, and Quality Controls
Privacy, Cost, and Quality Controls

The Result

  • 01

    Viewers can move from a spoken political claim to a clear verdict, supporting evidence, and source context without leaving the video.

  • 02

    The product uses AI to organize and explain evidence while keeping verdicts connected to sources instead of presenting the model as the authority.

  • 03

    Browser extensions, accounts, entitlements, billing, sharing, and session history turn the underlying verification platform into a usable consumer experience.

  • 04

    Reconnect handling, MV3 lifecycle recovery, graceful degradation, and observability create a stable foundation for real-time verdict delivery.

  • 05

    Cost-aware AI routing, telemetry boundaries, authentication, and operational monitoring support sustainable scale without compromising user trust.

  • background image

    Full case study

More Case Studies

Desktop background for section 'Let’s build what’s next'Mobile background for section 'Let’s build what’s next'
Label for proposal 'Let’s build what’s next'

Let’s build what’s next

Discuss My Project