A multi-agent decision engine and explainable-AI suite that optimizes Super Bowl advertising strategy — predicting ad effectiveness, modeling telecast audience flow, and quantifying the multi-million dollar ROI of a single spot. Built on a two-stage RidgeCV + Monotonic EBM prediction stack, vector RAG over marketing research, and a Google ADK boardroom of collaborating agents on Vertex AI, surfaced through an interactive Streamlit dashboard.
What it does
- Two-stage decision engine — RidgeCV + Monotonic EBM models predict ad effectiveness with interpretable, monotonic feature contributions.
- Vector RAG — Gemini embeddings (
gemini-embedding-001) retrieve grounded context from market research for strategy recommendations. - Multi-agent boardroom — Google ADK agents on Vertex AI debate, refine, and reach a collaborative ad-strategy decision.
- Market signal pipeline — yfinance equity returns, Google Trends search spikes, Wikipedia pageviews, and an event-study abnormal-return model (CAR[-1, +5]).
- Interactive Streamlit dashboard — explore advertiser performance, audience flow, and ROI predictions directly in the browser.
Live demo
← all projects