Own a production dbt project with 200+ SQL models structured across staging, intermediate, fact, and dimension layers, serving client-facing APIs for company, transaction, investor-profile, and country-risk products. Enforce data quality through Airbyte freshness validation macros, database constraints, and 35 dbt exposures documenting full data lineage across 7 product surfaces.

Designed dimensional models for high-volume financial and time-series data, including a country-risk warehouse covering daily, weekly, monthly, quarterly, and annual macroeconomic indicators across 15+ African countries. Used star-schema patterns with surrogate keys and precomputed aggregates (revenue bands, EBITDA margins, transaction history), all materialized as indexed tables to eliminate runtime join latency and support millions of API reads with precision.

Build and maintain 44 orchestrated Prefect 3.x pipelines processing end-to-end ELT workflows. Airbyte syncs ingest 61+ raw tables from operational sources into PostgreSQL, dbt transforms them into production-grade models, and downstream services (Qdrant semantic search, Algolia indexing, S3 documentation exports) consume the output. Implemented modular, reusable ELT patterns standardized across 20 domain modules with staged/production environment separation, enabling rapid onboarding of new data products without deep technical ramp-up.

Built a reusable AI-driven data extraction subsystem using CrewAI and Claude that parses unstructured sources (PDFs, HTML, APIs) into Pydantic v2-validated datapoints across 19 specialized flows (CPI, inflation, FX, public debt, monetary policy). This enables Finance teams to receive structured, compliant financial data without manual processing.

Led an end-to-end migration of the operational data layer from Airtable to NocoDB, rebuilding 199+ SQL models and re-architecting all Airbyte ingestion connections. This eliminated thousands in recurring tool costs while maintaining zero downtime for downstream financial data products

Engineered an IoT data pipeline using MQTT and Python (Pandas, Polars, NumPy) to ingest real-time streaming sensor data and large-scale historical weather datasets, preparing inputs for MIRNet — a deep-learning temperature-forecasting model later published in the FUOYE Journal of Engineering and Technology.

Developed preprocessing and feature engineering with Scikit-learn, optimizing data flow to an Nvidia GPU cluster for TensorFlow/Keras model training and supporting end-to-end data science workflows from ingestion through deployment.