AI-Institute-Africa/ChiedzaAI_Zivadzidzo

# ZivaDzidzo — education intelligence for school leaders ZivaDzidzo turns approved school-level context into practical curriculum, staff-readiness, and cohort-outcomes decision support for the OpenAI Build Week Education track. It is intentionally **LLM-native**: Gemini 3.5 Flash is the default production route, with a checked JSON schema. A backend operator may explicitly select OpenAI or Anthropic for the three structured assessment heads and tool-enabled Assistant; every provider response is still validated by the same server-side JSON/Zod contracts. It does not train, serve, or claim a predictive machine-learning model. ## What is built - A responsive Expo mobile/web workspace with five primary destinations: Dashboard, My School, Assess, Reports, and More. - An aggregate-only school-leader dashboard with readiness coverage, priority alerts, curriculum status, learning-outcomes trends, and recent activity. - Three structured assessment heads: - Teacher Roles — AI-disruption exposure and reskilling priority from staff readiness signals. - Learning Outcomes — cohort/subject-level pass-rate resilience; learner identifiers are rejected before any model call or persistence. - Curriculum & Future Skills — a Skills Obsolescence & Readiness Index (SRI), with the final score independently calculated on the backend. - Department and subject hierarchy, guided roster-import preview, confirmation, validation feedback, and an atomic database import function. - Private DOCX/PDF reports with executive summary, evidence, chart, timestamp, model/prompt version, actions, and caveat. - Role-filtered priority notifications that deep-link to the aggregate Dashboard. ## Trust and data boundaries - Gemini 3.5 Flash is the default/primary provider, configured in `backend/config.js` for both structured assessment and the tool-enabled Assistant. OpenAI and Anthropic are optional, backend-only providers selected by `LLM_PROVIDER`; there is no silent failover. - Scores, confidence, cav …