Built in the Deep End. Proven Across the Board.

The Challenge

A client needed an intelligent document retrieval system capable of answering domain-specific questions accurately — without hallucination — over a large proprietary knowledge base.

The Solution

Designed and built a production RAG pipeline with vector store indexing, semantic chunking, and retrieval-augmented generation. Implemented a modular architecture allowing the client to swap embedding models and LLM providers without rebuilding the system.

The Result

Delivered a fully operational AI assistant with measurable retrieval accuracy — replacing hours of manual document search with sub-second contextual answers.

The Challenge

Gokada — Nigeria's largest last-mile delivery service — faced a critical operational problem: drivers were positioned suboptimally across Lagos, resulting in high rates of unfulfilled delivery requests and lost revenue for both the company and its driver partners.

The Solution

Applied causal inference methods to identify the root causes of unfulfilled requests, then built a logistic optimisation model recommending strategic driver placement across city zones — using historical trip data, driver behaviour patterns, and geospatial analysis.

The Result

Produced a data-driven placement strategy with quantified improvement projections for fulfilment rates, directly tied to driver income and platform revenue.

The Challenge

A telecommunications company had extensive customer and network usage data but no systematic way to extract business insight — leaving strategic decisions based on intuition rather than evidence.

The Solution

Built a full analytics pipeline from raw data ingestion through to an interactive Streamlit dashboard — covering customer segmentation, usage pattern analysis, and operational KPIs — giving the analytics team a single, live source of truth.

The Result

Delivered an operational analytics dashboard enabling the business team to self-serve data questions, reducing analysis turnaround from days to minutes.

The Challenge

Rossmann Pharmaceutical needed to forecast sales six weeks in advance across hundreds of store locations — accounting for promotions, competitor proximity, school holidays, and seasonal patterns — to optimise inventory and staffing.

The Solution

Built an end-to-end ML and deep learning forecasting system using Scikit-learn, TensorFlow, and feature engineering pipelines. Trained models on multi-year historical data with full MLOps: reproducible pipelines via DVC and automated retraining workflows.

The Result

Produced a six-week sales forecast with strong predictive accuracy across store types, enabling confident inventory planning and reducing over/under-stocking.

The Challenge

An organisation needed a scalable, automated data pipeline to extract data from multiple heterogeneous sources, load it into a centralised data warehouse, and transform it into clean, analyst-ready tables — with data quality validation at every step.

The Solution

Architected an end-to-end ELT pipeline using Apache Airflow for orchestration, dbt for in-warehouse transformation, Great Expectations for automated data quality checks, and Redash for business intelligence dashboards on top of PostgreSQL.

The Result

Delivered a fully automated, observable data pipeline — reducing manual reporting effort and giving analysts reliable, validated data on a scheduled refresh cycle.