damn, so good to see this. Thank you, enjoy!

thank you, you should try it again, open settings (top right corner), click on the second option "Keywords".

i didn't want to make the onboarding screen more complicated, that is why is slightly more hidden in the settings.

Veganism spans so many fields that it works better as a keyword than a section. Add “vegan”, “plant-based” or something as keywords, and matching papers come first from every field the app covers: climate, public health, biology, law and more. I looked into it, and PhilArchive should be possible to add, so I’ll look into adding a Philosophy field.

Thanks, good suggestion! I looked into it and it should be quite easy to add. IACR ePrint has an open feed with abstracts and categories, and the metadata is free to use, so it fits the app.

Thanks, really useful feedback! I’ll look into the missing sources from that issue. It could well be a problem for environmental science and ecology too, since I’m not too familiar with those fields. If you know or find any preprint servers people in your line of work use, feel free to add them to the issue, that would help me a lot. A News tab is an interesting idea, I’ll think about it. I do want to keep it lean and focused on papers, so it would only make sense if it fits without making the app heavier or less private. I also really like how you’re using it. Starting with broader keywords and then narrowing down is something I didn’t think of, but I’m happy you tried it and made it work!

I hope you find it useful! Thank you

Wow, great! Thanks for using it! No, there’s no Gemini or any cloud AI. It’s a classic logistic regression model, all inside the app. I tested it against MiniLM, and it did just as well in my tests but would add like 35 MB, so I left it out. I also only build native Android apps, which stay much leaner than apps built with heavier cross-platform frameworks. A lot of apps could also be 10 times smaller if they didn’t include all the trackers and analytics. More details about the model: https://aftergleam.app/how-it-works

I appreciate it a lot, thanks!

oh yes that works, but I would recommend instead to use the search to find specific papers and upvote them to tune the classifier faster. Another way is to import a list of papers from your related work in .bib or .ris file. I will increase the keyword limit, can easily be more, i just didn't think it was needed. Thanks a lot! And for the feedback!

Amazing! Thank you

Oh and forgot to add, you can tweak the algorithm in the settings to put more weight on venue.

Not everything, no. All papers come from preprint servers like arXiv and bioRxiv, but many are posted after they’ve been accepted. Aftergleam shows that on the card and ranks accepted papers higher. If a paper you skipped gets accepted later, it can show it to you again.

Thank you so much for taking the time to give me your feedback! I have already made the it remember your scroll distance, but only on the For you page:) . I forgot to do the same for search results, so that will be an easy fix. Thanks for letting me know. As for the font size, it differs quite a lot between devices, so I think an option to adjust it in the settings will work.

Thanks! I agree, it’s not optimal for reading papers, but its great for keeping up in a specific field and finding new papers, which takes a few minutes a day, on the toilet. You can then, save it, share it, read it properly somewhere else. And for reading on the phone, there’s a reader mode for arxiv papers, so the text fits the screen and you’re not pinching around a PDF. Also, native android development is what I know best. I’ve made several smaller apps, and this idea had been brewing for a while.

huh, that's a quick verdict, let me explain. I'm glad you like the idea! It doesn't only pull from arXiv, it reads 8 preprint servers. The value comes from recommendations made by a small classifier trained on your phone from what you read or upvote/downvote (tf–idf and logistic regression). Yeah the docs are ai summaries and I included them in the repo. You can read how the algorithm works and what different things I tried in the "The model" section here: https://aftergleam.app/how-it-works

Not really. The humanities rarely use preprint servers, so there’s little to draw from. It covers computer science, physics, mathematics, statistics, engineering, economics and finance, biology, medicine, chemistry, psychology, social science, education and law, 114 subjects in all.

How I use AI in making it is transparent here and same rule applies for contribution: https://aftergleam.app/ai-notice

Lets also not forget that these booking UIs replaced a human operator and offloaded the work and time on you, the consumer to find yourself correct tickets. Each priced algorithmically as much as they can squeeze.

for every good ui there exists ten fold as many bad ui websites filled with ads, spyware and popups, I'm okay not visiting those sites myself. Of course you have to use local models if you wish to not get surveiled and analyzed by cloud ai providers.

Thanks for the kind words! To be fair, Syncthing itself is still actively maintained (they just released 2.0), it was the official Android app that got discontinued due to Google Play issues. Community forks still exist on F-Droid though.

But yeah, Syncthing and Synchi have different workflow. Syncthing needs daemons on all devices and can't sync to a mounted drive, NAS path, or local folders on the same machine. Synchi is on-demand and doesn't care where the two roots are. This is also why I started working on it. I used syncthing for a few years before that.

Totally fair, Unison is solid and I was inspired by it a lot, but its also the reason why I started working on my own version. I was frustrated with it because on Synology/SMB filesystems it kept seeing changed timestamps as modified files, so I'd constantly get fake changes and transfers. Synchi only treats a file as changed if the content hash actually differs.

Also, I once managed to delete my entire home folder because I tried syncing it with Unison to a server and it synced the empty remote folder back to my home, wiping everything. That's exactly the kind of human error Synchi tries to prevent by being very explicit and verbose about what it's about to do.

Thank you for keeping an eye on it!