[GSOC 2026] LLM Executor Prototype Feedback

[gauravdev95]

Hi @danselman @DianaLease,

I am a GSoC 2026 applicant interested in the “LLM Based Template Logic Executor” idea.

I explored the TemplateArchiveProcessor and built a working prototype where I replaced the TypeScript-based trigger() logic with an LLM-based executor (using Gemini). The prototype successfully generates state, response, and events from input data.

I wanted to ask:

1. Does this approach align with the expected direction of the project?
2. Should the LLM executor act as a fallback or primary execution method?
3. Are there specific parts of the codebase you recommend focusing on?

I would really appreciate your feedback.

Thanks!