About

About / Project Description:
This project answers the core hackathon question: Which user behaviors, workflows, and patterns on Zerve are most predictive of long-term success?
We define "success" uniquely as Zerve Virtuoso status — users who exhibit sustained high-value engagement:

Retention ≥90 active days OR upgrade to Pro
Workflow sophistication (≥3 deployments + multi-step complex sequences)
Platform contribution (collaborations, shared canvases)

We built a comprehensive, fully reproducible pipeline inside Zerve using the official hackathon dataset (~409K events from 4,775 users).
Methodology highlights:

Deep EDA + cohort/retention analysis with Plotly visuals
Novel feature engineering: Markov-style event sequence transitions, NetworkX collaboration graphs (centrality/clustering), "Spark Moments" in first 3 events, AI recovery patterns, velocity metrics
Interpretable ML: HistGradientBoosting/XGBoost for binary virtuoso prediction + continuous Mastery Score (0–100 weighted formula)
SHAP interpretability, KMeans personas (e.g., Champions, Explorers), early cohort insights
Live deployment: "predict_virtuoso" API endpoint for scoring new user histories

Key surprising insights:

First-week "Spark Moments" (e.g., quick first deployment) predict 90-day retention 3x better than total events alone.
Collaboration graph centrality beats raw event count as the top predictor of upgrades.
Users who recover fast from errors show 2.4x higher Mastery Scores — highlighting resilience as a hidden driver.

Business impact for Zerve: Prioritize onboarding nudges for early deployments/collaborations, target high-centrality users for Pro upsell, and use personas to personalize retention campaigns.
Fully agent-assisted + deployed on Zerve — showcasing its AI-native strengths for end-to-end data science. Ready to copy, run, and extend. Let's turn more users into Virtuosos! 🚀