nomemientas

Political speech analyzer. Paste a YouTube video, a tweet, or an article — the AI strips the rhetoric and tells you what they really meant.

Status: Live
URL: nomemientas.org
Stack: Next.js, Vercel, Cloudflare Tunnel, local GPU, Nous Research API (qwen3.6-flash)

Overview

nomemientas ("don't lie to me") is a tool that analyzes political speech in real time. It extracts transcripts from YouTube videos, tweets, and news articles, then runs them through an AI model that identifies honesty levels, logical fallacies, vague promises vs concrete commitments, and emotionally manipulative language.

Built in a weekend as a pure agentic build. Shipped to production without Carlos reading a single line of code.

Features

• URL analysis — Paste a YouTube, Twitter/X, or article URL
• Free text — Or paste any text directly
• Honesty score — 0–10 rating based on specificity, accountability, and rhetorical depth
• Claim extraction — Each claim tagged as fact, promise, attack, or opinion
• Fallacy detection — Named logical fallacies with examples from the text
• Vague vs concrete — Separates empty rhetoric from verifiable commitments
• Omissions — What the speaker deliberately left out
• Share cards — PNG image with the score and translation, ready for social media

Architecture

Browser → nomemientas.org → Vercel
  └── /api/* → tunnel → local server
       ├── YouTube extraction (transcript API)
       ├── Twitter extraction (meta scraping)
       ├── Article extraction (cheerio)
       └── AI analysis (qwen3.6-flash via Nous)

The frontend is served by Vercel's global CDN. API requests are proxied through Cloudflare Tunnel to the local machine where extraction and inference run.

Security

• Rate limited (5 req/min per IP, 100/day total)
• SSRF protected (IPv4, IPv6, cloud metadata blocked; redirects validated per hop)
• CSP headers
• Prompt injection hardened
• No user accounts, no login, no data collection