Yassin El Kaisi

Designed and developed the backend architecture for a multi-agent AI platform launched to general availability for over 4,400 employees, enabling efficient AI-driven information retrieval across the organization. Built a custom multi-agent framework integrating GPT-4o, GPT-5, and GPT-5-mini; consistently achieved higher retrieval quality than third-party agent frameworks on proprietary data. Developed a custom indexing engine written in Go to power multi-source knowledge bases. Automated agent creation workflows, improving deployment scalability. Integrated the Model Context Protocol (MCP) for interoperability and extensibility. Expanded data source support, including Confluence and other internal repositories.

Researched, designed, and implemented a state-of-the-art information retrieval system from scratch for internal proprietary data. Integrated cutting-edge LLMs (GPT-4o, GPT-5, GPT-5-mini) to improve reasoning and retrieval quality. Experimented with locally deployed LLMs and embedding models. Designed, implemented, and deployed multiple agents to production, receiving highly positive end-user feedback. Built a QA dataset generator and testing engine for randomized agent evaluation on private data. Leveraged vector, SQL, and NoSQL databases to optimize retrieval and handle diverse data structures.