A palestra intitulada “Knowledge Graphs + AI: The Evolution of Automated GitHub Issue Resolution” será apresentada pela Prof.ª He Ye no dia 17 de novembro, às 14:30, na sala B008, e será moderada pela Prof.ª Alexandra Mendes (DEI).
Resumo:
“AI coding agents are becoming increasingly capable, achieving strong results on benchmarks such as SWE-bench. However, most still struggle with real-world challenges such as issue reproduction, precise context retrieval from large codebases, and the high cost of LLMs. In this talk, I will introduce our recent code agent, Prometheus — a knowledge graph-powered, multi-agent system designed to tackle GitHub issues in practice. Prometheus transforms entire repositories into a unified knowledge graph stored in Neo4j for scalable and structured reasoning. This enables precise, cross-language context retrieval, allowing large language models to generate accurate and efficient fixes. Prometheus delivers robust performance, resolving diverse issues across seven programming languages. I will show how combining LLMs with knowledge graphs can advance automated issue resolution beyond today’s benchmark-driven limits. We have recently transitioned this research into an off-the-shelf product that helps industry resolve software issues automatically.”
Sobre a Palestrante:
He Ye é Professora Auxiliar na University College London. Anteriormente, trabalhou como investigadora de pós-doutorado na Carnegie Mellon University e obteve o seu doutoramento no KTH Royal Institute of Technology. A sua investigação centra-se no desenvolvimento da próxima geração de agentes de código para automatizar tarefas de engenharia de software, com foco na recuperação do contexto da base de código, resolução automatizada de problemas e construção da memória do agente de código. Além da academia, ela é cofundadora da EuniAI, uma startup comprometida em transformar pesquisas em soluções reais que ajudam os programadores a enfrentar desafios práticos de software.