Contrasta
Verifiable context for every claim. It finds the checkable statements in a speech and holds them up against official sources.
An engine, an API and a web playground. It never says "this is a lie". It says whether a claim is supported, partly supported, contradicted or not backed by enough evidence, and shows its sources and limits.
Built with
TypeScript · Node.js · React · Vite · AssemblyAI
What it is
Public debate is full of figures, laws and comparisons that are hard to check in real time. Contrasta picks out the verifiable claims in a speech, checks them against official sources and shows the context, the sources and what it couldn't confirm.
It's aimed first at Spanish politics, the media and the people who follow them. It doesn't judge people: it checks claims.
How it works
The engine is a pipeline of small, testable stages rather than one giant prompt: find the claims, classify them, search for evidence, propose a cautious verdict and keep an auditable trail of how it got there. A transparency mode shows that trail for every claim.
The AI provider can be swapped, and the whole thing runs with a mock provider and no API keys at all. The engine came first on purpose: the browser extension that listens to a video and contrasts what's said will be just one more client.
- TypeScript · Node.js
- React · Vite (playground)
- AssemblyAI (transcription)
Development
Phase 3 closed: evidence quality.
Guards against generic evidence, checks on how recent it is, and verifiers.
Phase 0: engine, API and playground.
It works end to end with a mock AI provider, without any API key.