Primordia Labs

Ground truth intelligence powered by Google Earth Engine (VIIRS satellite imagery) and Gemini (news sentiment). See the divergence between physical reality and market narrative before markets react.

Comment

Inspiration

Markets move on narrative long before reality is fully understood. Headlines, sentiment, and social amplification often drive price action faster than the physical world can respond—ports fill up, cities light up, supply chains move, but markets react to perception first.

We asked a simple question: what if markets could be grounded in reality again?

Primordia was inspired by the idea that economic truth is fundamentally multimodal. Physical signals from the planet and narrative signals from media should be analyzed together, not in isolation. By combining satellite-derived measures of real-world activity with Gemini 3’s reasoning and large-scale sentiment understanding, we built a system that reveals where perception diverges from reality before that gap closes.

What It Does

Primordia is a ground-truth intelligence platform that uses Gemini 3 as its analytical core to compare what is happening with what is being said.

It:

  • Pulls VIIRS night-lights data from Google Earth Engine to measure real-world economic activity across major global regions (Shanghai, Shenzhen, Suez Canal, Port of LA, Rotterdam)
  • Uses Gemini 3 to analyze global news streams, extracting sentiment, confidence, dominant narratives, and emerging risk factors
  • Computes a Divergence Index (0–100) that quantifies mismatch between physical activity and narrative sentiment
  • Surfaces signals when reality and narrative diverge, such as rising infrastructure activity amid negative headlines

Primordia turns fragmented signals into a unified, interpretable view of the global economy.

How We Built It

Backend (FastAPI)

  • Google Earth Engine integration for satellite time-series data
  • A Gemini-powered SentimentService built on Gemini 3, designed for structured, machine-readable outputs
  • Divergence logic combining physical signals and AI-derived narrative signals

Frontend (Next.js)

  • Interactive Mapbox-based global map
  • Region selector and signal cards
  • AI insight panel driven directly by Gemini 3 outputs
  • TradingView charts for financial and market context

AI Pipeline (Gemini 3)

  • Gemini 3 ingests batches of headlines per region
  • Applies multi-document reasoning to score sentiment, identify dominant themes, and flag systemic and regional risks
  • Returns consistent, structured JSON optimized for downstream analytics rather than chat responses

Challenges We Ran Into

  • Configuring Google Earth Engine authentication across local and cloud environments
  • Prompt engineering Gemini 3 to reliably return stable, parseable structured outputs
  • Managing Earth Engine rate limits using caching, aggregation, and graceful degradation strategies

Accomplishments We’re Proud Of

  • Live Earth Engine satellite integration with robust fallback modes
  • Gemini 3–driven reasoning and sentiment analysis with structured outputs including themes, risks, and confidence
  • A novel Divergence Index quantifying physical reality versus narrative perception
  • A production-ready architecture with caching, error handling, and cloud-friendly deployment

What We Learned

  • How to use Google Earth Engine for large-scale spatial and temporal economic analysis
  • How to leverage Gemini 3’s reasoning capabilities to transform unstructured global news into reliable structured intelligence
  • That narrative and physical signals frequently diverge, and exposing that gap creates real decision-making value

What’s Next for Primordia Labs

Primordia’s next phase is to evolve from a signal platform into a foundational AI system for real-world economic understanding, built natively on Google’s AI and geospatial stack.

  • World model for economic activity
    Use Gemini 3 to fuse satellite data, news, trade flows, and policy signals into a continuously updated economic world model that reasons about causality rather than correlation.

  • Multimodal planet-scale reasoning
    Expand beyond night lights to include NDVI, ship detection, mobility, and infrastructure change, enabling Gemini 3 to reason jointly over imagery, text, and time-series data.

  • Predictive narrative drift detection
    Train Gemini 3 to identify early narrative shifts when sentiment begins diverging from physical signals, allowing institutions to anticipate repricing, supply-chain stress, or geopolitical risk.

  • Vertex AI–powered vision intelligence
    Integrate Vertex AI Vision for automated vessel, vehicle, and industrial activity counting, turning raw imagery into structured economic primitives at global scale.

  • Gemini-native APIs for institutions
    Expose Primordia as a Gemini-powered intelligence API, enabling natural-language queries such as:
    “Which regions show rising physical activity despite negative media sentiment?”

  • Public-good and resilience signals
    Partner with researchers and NGOs to publish open indicators for disaster recovery, infrastructure resilience, and economic rebuilding, using Gemini 3 as the reasoning layer behind transparent metrics.

With Google’s AI, cloud, and geospatial infrastructure, Primordia has the potential to become a new class of global intelligence system that grounds markets, policy, and research in reality rather than narrative.

Built With

Share this project:

Updates