At TushindeBet, I built a fraud detection and reconciliation engine that handles uncredited deposits and unmatched transactions. It cut customer care calls about missing funds by 50% while processing about 1.5 million transactions a month.
I added an XGBoost risk model that reduced manual review volume by about 40% while keeping false positives under 2%. Reviewers’ feedback informs retraining every two weeks, and GPT-4o decision-support workflows automated about a third of manual case reviews.
At Lottomotto (Ebet Kenya), I built Node.js and TypeScript services for financial transactions and designed RabbitMQ event pipelines. I also implemented transaction state machines with idempotent guarantees and optimized PostgreSQL queries and connection pooling.
At TripleP TV (PPP), I built automation pipelines that reduced daily manual data processing from about four hours to roughly one, and APIs that cut report generation from about two days to under 30 minutes. Earlier, I developed Android applications at Ngomma VAS and taught Python, SQL, and machine learning at Ubunifu College.
