What the World Really Thinks

Explore global perspectives on any question using AI-powered analysis and Voice Commands

Comment

What the World Really Thinks

Talk to your dashboard and watch the world's opinions render in real time — AI-powered global perspective analysis, built on Amazon Aurora PostgreSQL and Vercel.

Have you ever talked to your dashboard?

Most people haven't. Now imagine asking "What does China really think about climate change?" out loud — and watching a live map analyze, think, and answer across dozens of countries in seconds. That's What the World Really Thinks: a voice-driven, AI-powered explorer of global opinion, with a Vercel front-end shipped in minutes and an Amazon Aurora PostgreSQL back-end designed for scale.

This is the "Zero Stack" in action: front-end in minutes, back-end built for millions.

Inspiration

The same question gets radically different answers depending on where in the world you ask it. Cultural context shapes belief, yet we rarely see those perspectives side by side. I wanted a tool that could instantly reveal global consensus and cultural divides on any topic — just by talking to it.

This project began life as "Ask the World Anything," which won Best Use of Actionable Analytics at the Tableau Hackathon 2025. For H0: Hack the Zero Stack with Vercel v0 and AWS Databases, I re-architected it from the ground up — moving the data layer onto Amazon Aurora PostgreSQL (Serverless v2) and shipping the whole experience on Vercel — to prove the concept can scale from a single dashboard to a global, production-grade application.

What it does

What the World Really Thinks combines the Perplexity API, ElevenLabs Voice Agents, Amazon Aurora PostgreSQL, and Tableau into a conversational analytics experience.

Speak a question like:

  • "Analyze: Social media makes people less connected"
  • "What do the United States, Canada, and Mexico think about vaccines?"
  • "Analyze climate change in China, India, and Japan"

The system:

  1. Listens to your voice via ElevenLabs Conversational AI.
  2. Extracts the question and the country names from natural speech.
  3. Analyzes each country's stance with the Perplexity Sonar model.
  4. Writes the results to Amazon Aurora PostgreSQL.
  5. Auto-refreshes the Tableau dashboard over a Live connection — no clicks, no extracts.

You get an interactive world map colored by agreement, a Global Truth Index (0–100), per-country stance scores, percentage breakdowns, and the cultural factors that explain why each country holds its view.

How I built it

The AWS database: Amazon Aurora PostgreSQL (the scalable backbone)

The heart of the H0 build is the data layer. Every voice or button-triggered analysis runs through a Vercel serverless function, calls Perplexity, then writes a wide-format result set into Amazon Aurora PostgreSQL (Serverless v2):

  • Designed for scale — Serverless v2 auto-scales compute with demand. Today it powers one dashboard; the same cluster scales to millions of "global perspective" queries with no re-architecture.
  • Live, zero-latency analytics — Tableau connects to Aurora over a native PostgreSQL Live connection, so a voice-triggered analysis appears on the dashboard the instant it's written.
  • Managed & secure — multi-AZ subnet group, automated backups, and TLS-encrypted connections, with zero database servers to manage.
  • Data model — a world_perspectives_sample table (one row per country) holding the question, stance, 0–100 score, agree/mixed/disagree percentages, three cultural factors, and geo coordinates.

Architecture: Voice → ElevenLabs → Vercel serverless function → Perplexity (Sonar) → Amazon Aurora PostgreSQL → Tableau (Live).

Backend (Node.js / Express on Vercel)

  • Vercel serverless functions with a 60-second timeout for long-running AI operations.
  • Perplexity API (Sonar) — culturally-nuanced, structured JSON analysis across up to 52 countries.
  • pg connection pool to Aurora, writing inside a transaction (clear previous analysis, bulk-insert the new one).
  • ElevenLabs webhook endpoint (/api/voice-analyze) that runs the analysis synchronously and reports status.

Frontend

  • Tableau dashboard + Extensions API for the interactive map and auto-refresh.
  • ElevenLabs Conversational AI widget that extracts the question and country names via LLM.
  • Dual-window status sync — voice popup and main dashboard show "Analyzing…" → "Complete!" together, coordinated with window.opener.postMessage().

Challenges I ran into

  1. Re-platforming the database for H0. The original ran on MySQL, which isn't an eligible H0 database. I migrated the data layer to Amazon Aurora PostgreSQL, converting the writer to the pg driver and PostgreSQL parameterized inserts. To keep it honest, I extracted the SQL builder into a pure function and added a unit-test suite (18 tests) covering placeholder math, multi-row inserts, and the data transforms — then verified the full Vercel → Aurora write path in production.
  2. Tableau API context limits. The Extensions API only runs in the main window. I bridged the voice popup to the dashboard with window.opener.postMessage() to trigger refreshAsync().
  3. Voice webhook timeouts. AI analysis takes 30–60s; ElevenLabs timed out at 20s. I extended the webhook timeout to 60s and made the endpoint synchronous.
  4. Country name → code mapping. Users say "United States"; the model needs "US". A normalization layer handles any capitalization, spacing, or full-name/abbreviation variation.

Accomplishments that I'm proud of

  • A conversational Tableau dashboard re-architected for the cloud — voice in, AI analysis, Aurora-backed, auto-refreshing out.
  • A deliberate, production-grade AWS data layer — Aurora Serverless v2 with a Live Tableau connection and transactional writes, verified end-to-end in production.
  • Engineering rigor — a pure, unit-tested INSERT builder and an 18-test suite guarding the migration.
  • Pushed Perplexity to return consistent, structured cultural analysis across up to 52 countries.

What I learned

  • Voice changes everything — adding conversational input fundamentally transforms how people interact with data.
  • The database choice is a design decision, not plumbing — moving to Aurora Serverless v2 turned a hackathon prototype into something that genuinely scales.
  • Prompt engineering is software engineering — structured, reliable LLM output takes the same rigor as code.
  • Truth is contextual — the same statement can be simultaneously accepted and contested depending on culture.

What's next for What the World Really Thinks

  • Historical tracking — store analyses over time in Aurora to watch global opinion shift.
  • pgvector semantic search — "show me past analyses similar to this one."
  • Multi-language voice — ElevenLabs supports 29 languages; analyze in users' native tongues.
  • Regional & comparative analysis — by continent, economic bloc, or side-by-side cultural factors.
  • Research & policy tooling — cross-cultural studies and initiative reception modeling.

Why this project deserves to win (mapped to H0 judging)

  • Technological Implementation — a deliberate Aurora PostgreSQL data model with transactional writes, a Live Tableau connection, Vercel serverless orchestration, and a unit-tested migration.
  • Design — a genuinely novel, hands-free conversational UX with synchronized multi-window status and automatic refresh.
  • Impact & Real-world Applicability — education, journalism, market research, and policy all need fast, explainable global sentiment; Serverless v2 makes it viable at scale.
  • Originality — a voice-driven, AI-analyzed, Aurora-backed global perspective engine — an evolution of an award-winning concept, rebuilt for the Zero Stack.

Built With

amazon-aurora · amazon-web-services · postgresql · vercel · node.js · express · perplexity-api · elevenlabs · tableau · javascript

Try it out

Built With

  • amazon-aurora
  • vercel
Share this project:

Updates