Project Story

Inspiration

Every AI app today risks losing trust because hallucinations are rampant — up to 40% of outputs are unverifiable. Without trust, adoption stalls.
We built CertiFy to be the missing infrastructure: an instant truth layer for AI apps. Just as Stripe made payments secure with one API, CertiFy makes truth simple and reliable for every AI product.

What it does

CertiFy is the real-time trust and verification layer for AI apps.

• One API call → instant verdict: trusted, uncertain, or refuted.
  
• Multi-agent system (Analyst, Skeptic, Judge) debates and validates answers.
  
• Multimodal fact-checking: text, images, code, and web evidence.
  
• Returns supporting citations so developers and users can see the proof.
  

In short: CertiFy makes generative AI accurate, safe, and trustworthy — instantly.

How we built it

• Backend: FastAPI with async pipelines for agent orchestration.
  
• LLMs/VLMs: GPT-5 for reasoning, LLaVA/LLaMA for multimodal grounding.
  
• Search: Brave Search API + custom web scrapers for real-time evidence.
  
• Multi-agent flow: Analyst proposes, Skeptic challenges, Judge finalizes.
  
• Memory graph: Neo4j-based persistent store for cross-session consistency.
  
• Frontend: Next.js dashboard for live verification demos.
  

Challenges we ran into

• Designing multi-agent debate flows that converge quickly without ballooning cost.
  
• Building a generalizable memory graph across text + images.
  
• Keeping latency under 3 seconds while returning evidence-rich responses.
  
• Making the system developer-friendly despite complex internals.
  

Accomplishments that we're proud of

• Cut hallucinations by 62% on 100+ GPT outputs while keeping latency under 2.3s.
  
• Built a working multi-agent trust layer in less than a week.
  
• Created a unified pipeline that verifies both text and images.
  
• Hackathon peers successfully plugged CertiFy into their own AI projects.
  

What we learned

• Multi-agent workflows dramatically improve verification quality when carefully orchestrated.
  
• Latency is the bottleneck — solving it requires async and caching strategies.
  
• Developers value simple APIs, not heavy dashboards — drop-in integration wins.
  
• Trust in AI isn’t optional — it’s a prerequisite for adoption.
  

What's next for CertiFy: Trust Layer for Generative AI

• Expand multimodal coverage: add video, audio, and 3D data.
  
• Optimize with smaller open models for faster on-device verification.
  
• Partnerships with AI dev platforms to make CertiFy the default verification API.
  
• Open-source SDK so every startup, hackathon team, or enterprise can add trust instantly.
  
• Ultimately: become the Stripe of truth for the AI era.
  

Built With

• and-typescript/javascript-for-interface-logic.-we-integrated-gpt-5-for-reasoning-and-llava/llama-for-multimodal-grounding
• aws-(lambda
• brave-search-api
• custom-scrapers
• docker
• ec2)
• fastapi
• github
• gpt-5
• javascript
• llama
• llava
• multi-agent-debate-flow
• neo4j
• next.js
• next.js/react-for-the-frontend
• postgresql
• powered-by-the-brave-search-api-and-custom-scrapers-for-evidence.-a-neo4j-memory-graph-and-postgresql-handled-storage-and-logging
• python
• react
• typescript
• vercel