Skip to content

GSoC Mentor Summit 2025

Posted on:November 4, 2025 at 12:00 AM

Starting this thread to share my experience of the GSoC Mentor Summit I attended last week. It was an overall great experience seeing people discuss OSS, their contributors, how their projects are going, and Zulip as well 🐙

There were some insightful sessions; some focussed around common problems faced by orgs during the program, and obviously many were around AI, its influence on the program (spam proposals). There were some sessions that were hosted by the GSoC admins.

GSoC Mentor Summit

Some of the sessions that I attended were:

1. Future of tech with AI

Folks talked about how AI is acting as a junior developer nowadays and how they are using AI for prototyping. But I felt the overall impression was that AI is still not at a point where it can just solve problems on its own in a way a contributor would do.

There was a segment about using RAG to improve the context for the used LLM which I personally think Cursor does a very good job at.

2. What to do if it looks like a project is going to fail?

There were talks about making sure the mentors are easily accessible. Some orgs had this rule that the contributor has to write a blog twice a week, telling what work they did during the time period. They also had a rule that you should make commits daily or communicate why that wasn’t the case.

Also, just letting the candidate know that the proposal is not something that is set in stone and is just a guiding document was also discussed, which I also feel is true.

Someone also said Google doesn’t allow people to participate who already have a job (not sure if that is true?).

3. GSoC Mentor Survey Results by admins

They talked about improvements to the GSoC dashboard some of them were:

One org does a final round of interview with the pre-final candidates and then pick the ones that they think performed well. Though that org had like 5 candidates to interview; but given we take ~13-15 candidates every year this might not be a feasible thing.

4. Changing roles/responsibilities for mentee/mentor in the AI era

Folks discussed identifying folks that show good signs and can be fostered early (even before they apply for GSoC) so that they can become successful contributors and in future maybe maintainers as well.

There was talk about mentees giving a final defense on their project based of which we can decide their evaluation results. I think this is just another way of ensuring contributors know what they are doing and not just asking AI do their work which results in them not learning (which then defeats the purpose of the program).

Also, the room was of the opinion that none of the things that a mentor does (motivating the mentee, tracking progress, giving feedback to the mentee, etc.) can be done by AI.

5. Fine-tuning an LLM on previously accepted proposals to build an AI reviewer

I was under the impression that this was about training/fine-tuning an LLM to help the admins filter out potential spam proposals. But this was more about helping the candidates review their proposals based on previously accepted proposals.

I think that the proposal’s content is not what gets it accepted; rather, it’s about the things the applicant does in the community (discussions, PRs, etc.).

Though they discussed using AI to examine whether a proposal has mandatory elements such as a project timeline, or any other thing that the particular org requires (this probably can be done through a chrome extension…?).

6. Tools and Techniques used by the community

Folks talked about tools that help them being productive or helped contributors. Some of them were:

The session notes have a complete list of the tools that were discussed.

Overall, it was a great experience attending the GSoC Mentor Summit 2025. It was nice to see people from the community come together and discuss the future of the program. It was also nice to see the admins taking the feedback from the mentors and working on improving the program.