Kola
Oyedele

Institutional reporting data arrived as zipped CSV archives dropped into SharePoint. There was no automated path from that drop zone into a governed analytics layer — which meant every reporting cycle involved manual downloads, manual unzipping, and manual loads, with no schema enforcement and no way to reconcile a bad load.

I built an end-to-end pipeline that authenticates to SharePoint through Microsoft Graph using a service principal, resolves the latest archive by matching a date pattern in the filename, moves it into the Lakehouse with a Data Factory copy activity in binary mode, then hands off to a PySpark notebook that extracts the archive into a curated landing zone and loads 12 CSVs into Delta tables. Load behavior is schema-aware per table: full overwrite where the source is a snapshot, append where it is an event log, and merge (upsert) where records can be restated.

The faculty evaluation dashboard was broken and had fallen out of use. Reporting had reverted to manual assembly, which was slow enough that the numbers were stale by the time anyone saw them.

I restored the dashboard end to end and enhanced it rather than patching it — rebuilding the model underneath, then adding complex T-SQL views for Heliocampus survey reporting on top. Departmental access is handled with dynamic row-level security bound to USERPRINCIPALNAME(), so a chair sees their department and only their department, from one published artifact.

Ten-plus departments each pulled their own numbers from SharePoint, Azure SQL and internal systems. Definitions drifted, retrieval was slow, and leadership could not compare figures across units with any confidence.

I led end-to-end data mapping for the SharePoint-integrated repositories with Dataflow Gen2, Power Query and SharePoint Lists, then built the dimensional models beneath a shared reporting layer. Automated validation with Dataflows and DAX logic caught bad data before it reached a dashboard, and I standardized data dictionaries and catalogs across 50+ critical datasets so a metric meant the same thing in every department.

Shipments were missing their dates and nobody could say why. Inventory and logistics data sat in SQL Server, but the reporting on top of it took eight hours a week to assemble and arrived too late to change any decision.

I extracted and transformed the SQL Server data with T-SQL and Power Query into inventory optimization dashboards, then automated the recurring reporting with SQL Server Agent and the Power BI Service so it produced itself. Alongside that I ran a governance pass that resolved 100+ data inconsistencies, which is what actually made the supply-chain forecasting trustworthy.

COVID and vaccination reporting had to be accurate, auditable and on time across 20+ clinics, drawn from clinical records that were never designed to be analyzed. Late or wrong submissions carried real financial penalties.

I extracted, cleaned and validated 2M+ patient records into tracking datasets that clinics and external agencies could both rely on, and optimized the SQL behind the operational dashboards until processing time dropped by half. Every federal and grant compliance report went out on time.

Raw sales CSVs with no model underneath them — the classic starting point. The goal was a pipeline that could scale from flat files to a warehouse a business could actually query.

I designed fact and dimension tables in Snowflake, built the load from raw CSV with Python, and served the curated model into Power BI for regional performance, product category and monthly growth reporting.