AI-Institute-Africa/ZivaBasa-MVP

# 🕴️🤔 ZivaBasa 😶‍🌫️ ### AI-Powered Workforce Intelligence Platform — MVP (Kaggle-Data Phase) **Module:** ZivaBasa (part of the ChiedzaAI platform — jobs, employment, productivity & skills forecasting) **Phase:** MVP prototype using public Kaggle datasets as a stand-in for real banking-sector data **Status:** Working end-to-end prototype (frontend + API + models) on proxy data — not a real-world findings phase --- ## 1. Purpose of This Phase This MVP validates the **explainable multi-task deep learning architecture** proposed for ZivaBasa (shared representation trunk → Employment / Productivity / Skills task heads → SHAP explainability layer), served through a real API and dashboard, before real Zimbabwean banking-sector data is available. **What this phase proves:** - The multi-task neural network trains and produces sensible per-task predictions - The feature engineering pipeline (raw → engineered → learned → fusion) works end-to-end - SHAP explainability runs correctly against a multi-output Keras model - The full stack — FastAPI serving predictions/explanations, a React dashboard consuming them, batch CSV upload, and an LLM-backed chat interface — works together as a real product, not just notebooks **What this phase does NOT prove:** - Anything about actual Zimbabwean bank employment/productivity/skills dynamics - Real predictive accuracy on the target population — Kaggle data is a **proxy**, not ground truth > ⚠️ Every dataset used here is a substitute for real banking HR/operational/AI-system data. > All findings from this phase are **methodological**, not empirical. This must be stated > explicitly in any write-up, thesis chapter, or stakeholder demo that references this phase. --- ## 2. What's Real vs. Prototype (read before demoing) - **Predict → Upload & Analyze** — real. Upload a CSV per task (Employment/Skills/Productivity), the backend matches columns by name, engineers features automatically, scores every row, and returns KPI cards, a departm …