GSoC 2026: Automated Agent System for Adding New Model Support to Optimum-Intel

[kyoui-dev]

Hello @popovaan and @rkazants,

I'm Kiho Kim. I'm passionate about Agentic AI, and got interested in LLM inference optimization after discovering FriendliAI last year. As the AI inference paradigm shifts from single-call to multi-step reasoning, I believe inference optimization has become far more critical in terms of cost and latency, and this is why I'm drawn to the OpenVINO project.

As an Agentic AI engineer intern last year, I built various agentic systems with LangGraph. Here are some of my relevant projects:

• Agent Builder Service — generates executable agentic workflows from user requests
• Property Submarket Analysis Workflow (link) — multi-agent collaboration using Agent-as-Tool pattern
• Research Paper Summarization and Analysis Workflow (link) — step-by-step summarization and analysis using prompt chaining

I'm interested in Project 12: Automated Agent System for Adding New Model Support to Optimum-Intel for GSoC 2026, and I have a few questions to get started:

1. Most of the good first issues on the board seem to be assigned already — could you assign one that would be suitable for a beginner preparing for this project?
2. I noticed several model support-related issues in the Optimum-Intel repo. Could contributing to one of those count as a prerequisite task?
3. Could you pick one of the recent PRs where model support was manually added? It would help me understand how model support is typically added and how to design the agent system accordingly.

Looking forward to your response!