Fact Pulse

• 

  How a spoken claim becomes a score.

• 

  Fact Pulse: ask Alexa+ if what you heard is true.

• 

  Fact Pulse remembers what you checked before.

• 

  A true claim scores 100, backed by ISRO and NASA.

• 

  Opinions aren’t scored, and Fact Pulse says so.

• 

  Every answer is a real call to our MCP server.

• 

  A myth scores 0, and Fact Pulse explains why.

Inspiration

Claims spread faster than anyone checks them, and a spoken answer from a voice assistant carries more weight than a search result because there is nothing to compare it against. We had already built Veridex, a web platform that scores how reliable a public claim is. Fact Pulse asks a simple question: what if you could ask Alexa+ about a claim and hear a short, honest verdict with the reasons behind it?

What it does

You ask Alexa+ whether a claim is true. Fact Pulse returns a 0–100 reliability score, a one-sentence verdict, and the main sources behind it, for example "62 out of 100, backed by two news reports but no official source." [Describe exactly what your demo does.]

How we built it

Fact Pulse reuses the scoring engine from Veridex. The score comes from 6 transparent rules: source-type weights, source diversity, freshness, a single-source penalty, moderator review, and an AI evidence check using Google Gemini. The AI can move the score by at most −30 or +20 points, so it adjusts the result without deciding it. Any claim with no official, document or media source is capped at 40. The backend is Node.js and Express with MongoDB Atlas, with a REST API of 30 endpoints. [Explain how Alexa+ connects to it: the skill or integration, and how a spoken request reaches the API.]

Challenges

A score that reads well on a screen is hard to say out loud. We had to cut the breakdown down to a verdict and a few sources that someone can follow in about ten seconds. [Add the real problems you hit, such as latency, speech recognition on claims, or sources that are hard to read aloud.]

What we learned

Keeping the AI's influence capped made the score easier to explain, which matters more when the answer is spoken. [Add what you actually learned from the build.]

What's next

[Languages, more source types, follow-up questions like "why that score?"]

Built With

• alexa+
• express.js
• google-gemini-api
• javascript
• jwt
• mongodb-atlas
• node.js
• render
• rest-api
• vercel