Spectraxis – Image Observability & Telemetry

Work-in-progress demo. Spectraxis ingests a set of creatives, extracts emotional signals (valence & activation), aggregates them into an Emotional Impact Score (EIS), and surfaces counterfactual suggestions and model receipts via a one-screen dashboard.

What & Why

• What: End-to-end demo of image observability—valence/activation estimates, Emotional Impact Score, counterfactual variants, and signed JSON receipts.
• How: CLIP embeddings (or demo mocks) feed tiny linear heads for valence/activation; scores funnel into EIS; a Streamlit dashboard and FastAPI endpoints expose results.

Status

• Streamlit app skeleton
• Demo data (20–50 creatives, fake engagement labels)
• Ingest/Featurize/Score pipeline (stub)
• Counterfactual optimizer (stubbed suggestions)
• Receipts writer & metrics trackers
• Tests for receipts/metrics/optimizer stubs
• README documentation and Loom demo

Quick Start

Coming soon.

make setup
make demo

Project Layout (planned)

spectraxis/
  app/
  data/demo/
  notebooks/
  receipts/
  tests/
  spectaxis/            # core package
    ingest.py
    featurize.py
    heads.py
    score.py
    optimize.py
    receipts.py
    metrics.py
    config.yaml

License

MIT License.