GSoC'26 introductory blog and author profile#850
Conversation
🎉 All Checks Passed!
✅ Completed Workflows
🚀 This PR is ready for review and can be safely merged to Great work! Your code meets all quality standards. 👏 |
cf9293e to
274661c
Compare
🎉 All Checks Passed!
✅ Completed Workflows
🚀 This PR is ready for review and can be safely merged to Great work! Your code meets all quality standards. 👏 |
274661c to
49e3455
Compare
🎉 All Checks Passed!
✅ Completed Workflows
🚀 This PR is ready for review and can be safely merged to Great work! Your code meets all quality standards. 👏 |
49e3455 to
3965063
Compare
🎉 All Checks Passed!
✅ Completed Workflows
🚀 This PR is ready for review and can be safely merged to Great work! Your code meets all quality standards. 👏 |
3965063 to
2f6acc6
Compare
🎉 All Checks Passed!
✅ Completed Workflows
🚀 This PR is ready for review and can be safely merged to Great work! Your code meets all quality standards. 👏 |
| title: "GSoC'26 Contributor" | ||
| organization: "SugarLabs" | ||
| description: "GSoC'26 Contributor at SugarLabs working on Sugar Activity on Demand" | ||
| avatar: "https://avatars.githubusercontent.com/u/Ashutoshx7?v=4" |
|
|
||
| ## Introduction | ||
|
|
||
| Hey, I'm Ashutosh ([@Ashutoshx7](https://github.com/Ashutoshx7)), a developer and artist. This summer I got selected for GSoC 2026 with Sugar Labs, and I'm working on a project called **Sugar Activity on Demand**. |
There was a problem hiding this comment.
I recommend that you put a link to the GSoC ideas page that references "Sugar Activity on Demand" and remove the bold.
|
|
||
| We talked about what "working" actually means. A generated activity doesn't need to be perfect. If it launches, shows a UI, and does roughly what the user described, that counts. We can always iterate from there. | ||
|
|
||
| We also had a good conversation about model selection. Walter's take was that we shouldn't just pick one model and run with it. Instead, we should benchmark a bunch of them against the same set of activity descriptions and compare the results properly. He specifically mentioned trying **Claude Opus** for its instruction-following ability. I also suggested testing some of the strong **Chinese models** like Qwen and DeepSeek, since they've gotten really competitive for code generation and in some regions where Sugar is deployed they might be more accessible or affordable. Walter was on board with that. So the plan is to build a model-agnostic evaluation phase before committing to any specific backend. |
There was a problem hiding this comment.
I recommend replacing "Claude Opus" and "Qwen" and "Deepsink" with some relevant roles and removing the bold.
|
|
||
| The offline question keeps coming up too. A lot of Sugar deployments have limited or no internet. Whatever we build needs a path toward running inference locally, maybe through `ollama` with a quantized model. It won't be as good as a cloud-hosted model, but a working offline mode is more important than a slightly better online one. | ||
|
|
||
| Walter also forwarded me an email from Gary Stager that really drove home why this project matters. Gary had uploaded a PDF of Brian Silverman's old Apple II program called *The Phantom Fishtank* (a cellular automata explorer for learners) to an AI tool, and a few minutes later had a working web version of software that had been lost to platform shifts decades ago. The original point was about how teachers now have a practical way to bring great educational software back to life, not by buying it but by making it. That's basically the thesis of Sugar Activity on Demand. If a teacher has an idea for an activity, working software should be within reach, not weeks of Python development away. |
There was a problem hiding this comment.
Please link to "The Phantom Fishtank". I think readers will benefit from that.

Description
Adds my author profile and introductory blog post for GSoC 2026
Type of Change
Accessibility
PR Checklist