Interest in Deep Search AI Assistant project (GSoC 2026)

[Lagmator22]

Hi @HongboZhao @KundaXu,
Interested in the Deep Search AI Assistant project for GSoC 2026.

Background:

• 3 PRs submitted to OpenVINO ([CPU] Fix TopK reference implementation to handle NaN inputs correctly #33633, [GPU] Fix ambiguous fmin/fmax calls for unsigned integer types in eltwise kernel #33667, Add f64 element type support to interpolate op, fixes issu #33570 #33572)
• Working with C++, learning Python.
• Familiar with APIs, RAG concepts and local LLM deployment.

Questions:

1. For multimodal data (PDFs, images, videos) should I focus on one format first or handle all together?
2. For the RAG pipeline are there any preferred OpenVINO examples or notebooks to start with?
3. Regarding AIPC deployment with GPU/NPU switching, should this be automatic based on workload or user-configurable? Haven't worked with NPU before so trying to understand the scope.
4. The expected outcome mentions "fuzzy search" is this semantic search via RAG or are you looking for something different like approximate string matching?

Planning to build a basic demo to get familiar with the stack. Any guidance appreciated.

Thanks

[adrianboguszewski]

Hi there, OpenVINO has not yet been officially announced as a 2026 organization. We don't know if we're accepted. Please wait for the official announcement.