khannyasha/deploysafe-zimbabwe

# DeploySafe Zimbabwe Institutional AI assurance and deployment decision platform. MVP v0.2.0 for the 2026 AI for Impact Challenge (AI4I), Track 4: Deployment. > **This repository contains a functional assurance and governance MVP developed for controlled sandbox validation. Demonstration data is synthetic. The system is not yet a production-certified platform, and no confirmed institutional pilot partner is claimed unless separately documented.** ## Problem Zimbabwean institutions are adopting AI-enabled systems faster than they can govern them. Two gaps recur: an AI risk visibility gap (risks exist but are not measured or documented) and an evaluation-to-governance gap (evidence exists but never becomes an accountable decision). Institutions need a practical way to measure fairness, local-language parity and drift, review security controls honestly, and record human deployment decisions with a defensible audit trail, all without production data, large budgets or specialist tooling. ## Solution DeploySafe Zimbabwe is a lightweight web application that carries one AI system through a complete assurance workflow: register the system, classify deployment risk with transparent rules, calculate Disparate Impact Ratio and Statistical Parity Difference (with bootstrap confidence intervals) from entered or uploaded data, record English, Shona and Ndebele evaluation results, compute Population Stability Index and KL-divergence drift with configurable alert thresholds, link alerts to fallback actions, review security controls against five explicit evidence states, define the pilot sandbox, and pass everything through a server-enforced human decision gate that produces stamped, exportable decision and audit records. ## AI4I Track 4 alignment | Criterion | What this MVP demonstrates | |---|---| | C1 Ethics and explainability (40) | Quantitative DIR and SPD with 95% bootstrap CIs and small-sample warnings; quantitative English/Shona/Ndebele parity with gap flags; plain- …