Workshop "Research in Action"
Abstract
This workshop brings together researchers and practitioners advancing the reliability, efficiency, and evaluability of LLM systems and agents in real-world applications. Speakers present procedural memory in LLM agents to enable reusable, adaptable skills across enterprise tasks; techniques for producing reliable answers; strategies for tabular data classification using LLMs; and a framework for automatically generating LLM-based judges from execution traces to streamline evaluation of agentic pipelines. Emphasis is placed on turning LLM capabilities into robust, measurable components through controlled skill transfer, uncertainty-aware decoding, data-efficient learning, and trace-driven evaluation, so that business can deploy agents with greater confidence, lower annotation cost, and clearer performance guarantees.
Speakers
Sber AI Lab
Sber AI Lab
Sber AI Lab
Sber AI Lab
ITMO University