Project, 2026.02–2026.12, in progress

Advancing an On-Premise Multi-SLM-Based AI Agent for VPD Control Optimization

The 2025 VPD AI agent prototype for LG Electronics relied on a single commercial general-purpose LLM API, which raised cost, data security, and availability concerns. This LG Electronics industry-academia project (Feb to Dec 2026) moves the agent to an on-premise multi-SLM architecture: several small language models, each fine-tuned with PEFT (LoRA) for one function, split the work of recommending simulation parameters and deciding when to stop iterating inside a LangGraph loop. The plan has three stages, extending next to adaptive context management and orchestration across multiple users and circuits. Joonho Seo is the project's lead researcher on the university side. The project is ongoing: the interim review took place in July 2026, and work is now on the second-half stages, adaptive context management and multi-user, multi-circuit orchestration.

Where

LG Electronics

My role

Working-level project lead

Status

ongoing