Veryfi

Inspiration

AI-generated content is becoming indistinguishable from reality. Deepfakes, synthetic voices, and fabricated news can now be produced at scale, powered by models from companies like OpenAI and Google. We were inspired by a simple but urgent question: what happens when video, audio, and online content can no longer be trusted? As misinformation evolves from false text to full reality manipulation, society needs a new layer of verification. We built Verify to help protect trust in digital information before it collapses.

What it does

Verify is a web application and Chrome extension that helps users quickly assess the credibility of online content including URLs, social media content, images, audio, and video to determine whether itis true, false, misleading, or unverified.

Web Application

Analyze Page: Upload text, images, or URLs to receive a detailed report with a verdict (true, false, misleading, unverified) and supporting reasoning.
Heatmap Dashboard: View a global map showing misinformation hotspots, trends, and category breakdowns.
Reports Page: Access the history of all past analyses.

Chrome Extension

Real-Time Scanning: Analyzes visible content on the current webpage and provides a quick credibility summary.
Highlight to Analyze: Select specific content for a deeper report powered by the full Verify system.

How we built it

Frontend: React 18 + Vite + TailwindCSS — web app with Analyze, Heatmap, and Reports pages
Backend: FastAPI + Pydantic v2 + Motor (async MongoDB) — handles scraping, AI agent debates, multimodal analysis, and report generation.
Database: MongoDB Atlas — stores users, reports, embeddings; supports vector search, geospatial queries, aggregation pipelines, and change streams.
Infrastructure and Deployment: Vultr + Docker Compose + Nginx — scalable, low-latency hosting for web, backend, and AI services.
AI & APIs: Gemini 1.5 Pro for deep analysis and AI agent debates; Gemini Flash for fast triage in the Chrome extension; integrated web scraping and fact-check APIs for source verification.

Challenges we ran into

During development, we faced several challenges. We quickly hit Gemini API rate limits, so we implemented throttling and caching to keep analyses running smoothly. Setting up the Vultr server was tricky, requiring careful configuration of networking and environment variables. Using Docker and Docker Compose allowed us to containerize all services backend, frontend, database, and Chrome extension ensuring reliable local and production deployment.

Accomplishments that we're proud of

AI orchestration is complex: Coordinating Gemini, CNNs, SVMs, and RoBERTa models requires fallback strategies and careful pipeline design.
Privacy matters: Minimal data storage, secure uploads, and transparent disclaimers are essential for user trust.
Power of MongoDB Atlas: Vector search, geospatial queries, and aggregation pipelines make real-time dashboards and trend analysis possible.
UX is crucial: Technical solutions need intuitive interfaces, like unobtrusive Chrome extension flags and clear report layouts.

What we learned

Built a multi-platform proof-of-concept: Web app, Chrome extension, and a demo real-time verification room.
Implemented core AI pipelines: Fact-checking with Gemini Agent Debate, multi-modal deepfake detection, and scam detection.
Real-time misinformation dashboard: Geospatial heatmaps, trending narratives, and live updates using MongoDB Change Streams.
User personalization: Preferences, categories, and historical reports stored and retrievable
Explainable verdicts: Each flagged claim shows AI reasoning, confidence scores, and supporting evidence.
Community layer added: Users can vote on credibility, submit evidence, and influence trust scores.
Hackathon-ready deployment: Fully containerized with Docker and Vultr-ready configuration for quick demos.

What's next for Veryfi

We would like to expand real-time detection capabilities, including live verification during video calls and instant alerts for trending misinformation on social media. Alongside this, we plan to harness the power of community intelligence allowing users to submit claims, vote on credibility, and provide feedback so that AI insights are enriched with human judgement, creating a dynamic and trustworthy verification ecosystem.