This week was the week for both mentors and contributors to turn in their evaluations. Mentors would give feedback on the experience with mentees, and mentees would do the same for mentors. Spoiler alert: I passed mine at the end of the week. π
Let's get right to my summary of the week.
Weekly Check-in
My lead mentor (Storm) was on holiday, so Saptak led the session. We started with a rundown of the state of the project so far. I mentioned that the PR for the new ImageBlock
was in good shape, I was making some good findings on the AI stretch goal, the image_description
still needed extra feedback, and my documentation (as well as the release notes) were also coming up nicely.
Saptak mentioned wanting to get the RFC moving faster so that we could get it approved/accepted by the core team, to meet the GSoC timeline. He mentioned he'd try to push it across to some other people to get more comments on it.
We agreed that for the rest of the sprint, I would ideally try to get the PR ready for review again, and I would also get a proof of concept ready for the AI stretch. One of the most exciting parts of my Summer of Code was about to commence. The agreed progression was to start the AI project as a regular Django project, and then modify it to suit Wagtail's needs iteratively. I was asked not to worry too much about the UI, but to get the basics up and running.
Accessibility Team Meeting
A few minutes before the Accessibility team meeting, I got a message from Thibaud. He had some concerns with the RFC and mentioned that he felt it required a bit more work. He wanted the RFC to be more detailed in covering the behavior for various use cases of the block, as well as how the changes would affect existing usage even when the block wasn't used. He mentioned the idea of developing a flowchart, which he eventually did with me after the Accessibility meeting, over a call. It was quite a long call, but I appreciated the time he put in, and we had a good discussion on what I could do to communicate this flowchart better in the RFC itself. It did mean I would have to go back to the drawing board, though.
What was particularly a debating point (which still hasn't been resolved) was whether Wagtail should keep the existing behavior (of defaulting to image titles) in the absence of any form of alt text. Image titles are meant to be brief, and are usually whatever the image was saved as in the computer (unless the user changes them). In lots of cases, though, the titles make very bad alt text. With no alt text, Wagtail's Accessibility Checker would easily flag a problem. With bad alt text, the Accessibility Checker wouldn't find it. At the point of my writing this, there are modifications in the works to improve the Accessibility Checker's ability to spot bad alt text patterns, but I'm working in parallel because I'm not sure how soon it will be ready.
Thibaud is of the opinion of falling back to titles, but my mentors and I are not. We will take it up with the accessibility team again, as well as the core team to gather overall thoughts. We will conclude in the coming weeks, though.
The meeting ended with some potential UI features we could integrate if the GSoC timeline permits. It also meant that I wouldn't be able to mark the PR as finally ready for a review till the pain points were settled.
Challenges
My biggest challenge came from trying to get the AI project prototype to work. This was largely because I didn't have any OpenAI access keys. I couldn't seem to get one from Thibaud earlier in the week so it halted my progress greatly. I set up the project template from the OpenAI docs I had read and tried to make sure that the only delays were from access keys.
I was able to get access keys from a friend of mine, but it had to be during his office break hours before invalidation. It was better than nothing, and I used this up until I got the keys from Thibaud during the Accessibility meeting. That was my major challenge of the week. The next challenge was (and still is) trying to account for the discussions in the Accessibility team meeting.
What I learned
I learned how to use OpenAI's API to take advantage of the multimodal features, I learned about some more use cases of images in models, and I learned how OpenAI works with images.
I'd say my best learning moment came from drafting the Flowchart with Thibaud. It showed me my lapses in knowledge of Wagtail's image management, and it also tested how confident I was in the work I had done so far. These things make all the meaning to me, and I'm glad to be able to learn more (even if it means constantly going back to the drawing board).
Evaluation
It was another great GSoC week, and I passed my midterm evaluations. I got great feedback, and I am excited about the remaining stretch of the program.
Thank you for reading.
Cheers. π₯
Top comments (0)