Synthetic SCADA telemetry from 19 utility-scale power plants across 7 African nations — ingested via Azure Databricks, transformed through a Medallion Architecture, and surfaced with 8 ML models and 111 Power BI measures.
Content-based nearest-neighbour similarity (cosine) over real plant attributes (technology, capacity, region) and real aggregated operations data (average availability %, average capacity factor % from 12,400+ daily SCADA rows) — not customer-purchase collaborative filtering, since a power-plant portfolio has no such history. Framed for ops triage: if one plant's numbers move, which plants have a similar profile and are worth a second look too. 19 plants is too few for a held-out accuracy claim, so this is reported as a diagnostic, not a validated prediction: each plant's nearest neighbour shares its primary technology 88.2% of the time.