Deep dives into AML transaction monitoring, KYC/KYB identity verification, computer vision, NLP for compliance, and building AI systems for Africa.
Most financial institutions still rely on static thresholds to detect money laundering. The result? Thousands of false alerts and sophisticated patterns slipping through. Here's why, and what ML changes.
Africa had no comprehensive open-source PEP database. We built one: 27,000+ profiles, FATF-compliant classification, hybrid fuzzy matching, and graph-powered relationship mapping.
Research papers report 99% accuracy. Production reality is different. The failure modes nobody publishes, the adversarial attacks customers actually try, and the decisions that matter.
Shell companies, nominee directors, circular ownership: the structures hiding UBOs are graph problems. Why SQL fails and how Neo4j makes KYB compliance tractable.
When 95% of your AML alerts are false positives, your compliance team isn't investigating, they're drowning. How we rebuilt a monitoring system from scratch.
How to match names across Arabic, French, English, and African language transliterations for PEP and sanctions screening at scale.
Five specialized AI agents that conduct full KYC/AML investigations autonomously: from alert triage to SAR drafting. The architecture, the guardrails, and what we learned.
500 million Africans lack formal ID. National databases are fragmented. Traditional KYC will never reach Africa's unbanked: only AI-first systems built for African realities will.
How llama.cpp enables running advanced AI language models on consumer hardware, no supercomputer required.
From test tubes to neural networks: how a career pivot led me to discover my true passion in artificial intelligence and machine learning.