Catching invisible errors: How I built a duplicate detection agent for Kenya's HIV program

Fredrick Kioko, a solutions architect in Nairobi, addressed the challenge of duplicate patient records within Kenya's HIV program by developing a multi-agent system using Elasticsearch during a hackathon. The system, built with Elasticsearch 8.11 and Elastic Agent Builder, comprises three specialized agents: a detection agent, a risk assessor, and an action recommender. This system efficiently scans records to identify duplicates through cross-facility pattern matching, demographic analysis, and temporal anomaly detection, and assigns risk scores to prioritize cases for healthcare workers. By providing specific, evidence-based explanations for flagged records, the tool enhances trust and usability among monitoring and evaluation officers, potentially saving significant resources and improving data reliability for treatment decisions. Following successful testing, the system is poised for pilot implementation in Nairobi and eventual national scaling, with projected annual savings of $195,000 and a 70% reduction in duplicate testing.