Project, 2025.08–2025.12

AI Agent Prototype Development for LLM-based VPD Control Optimization

In LG Electronics' simulation workflow, engineers had to change parameters and rerun simulations by hand, and the existing chatbot could only look up results. This industry-academia project addressed that along three lines: an LLM-based AI agent that takes natural-language instructions and loops through running a simulation, analyzing the result, and adjusting parameters on its own; a RAG system that grounds answers in internal documents; and a preliminary study on replacing external APIs with an on-premise LLM. Joonho Seo was the working-level project lead, responsible for carrying out the research overall and serving as the day-to-day contact with LG Electronics. The final report stated that all three KPIs were met: task time for the same simulation dropped by about 50% (from about 2 minutes to about 1), the RAG system reached a 91% hit rate and 91% MRR, and a benchmark of three open-source LLMs against GPT-4o confirmed the feasibility of a self-hosted LLM.

Where

LG Electronics

My role

Working-level project lead

Status

completed

3 KPIs reported as met (about 50% shorter task time, RAG hit rate 91%, local LLM benchmark)