Davit Gyulnazaryan
Reasoning Augmented Retrieval
Bio:
Davit Gyulnazaryan is an aspiring researcher and recent graduate of the American University of Armenia, working on AI systems for multimodal information retrieval. Davit is passionate about building practical AI tools to solve complex, real-world problems. His research interests include information retrieval, semantic search, ontology-based systems, knowledge graphs, multimodal question answering, and applications of machine learning in biology.
Description of the Talk:
This talk presents Activeloop-L0, a reasoning-augmented retrieval system for open-domain question answering on multimodal data. Unlike traditional RAG pipelines, it dynamically interleaves retrieval and reasoning, breaking down complex queries into sub-steps and refining results with vision-language perception. The system surfaces relevant passages from text, tables, charts, and diagrams, achieving state-of-the-art accuracy on visually rich benchmarks. The discussion will highlight the methodology, results, and future directions for building more reliable, tool-use-driven AI systems.