Could you please add https://eprint.iacr.org/ for cryptography?

The app works really well. Thanks for meantioning that it is vibe coded. I was looking for good ai products

Needs a section on veganism (the subject spans climate science, philosophy of mind, ethics, public health, law, biomedical engineering, etc)

OK, Sable is fucking amazing. there's like three other people on the FAF matrix that are using it now, thanks for the recommendation!

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.

Definitely not my line of work. I'm on the tools - https://www.explainxkcd.com/wiki/index.php/2501:_Average_Familiarity

The person that wrote the github issue seemed like they were somewhat aware of sources. Perhaps you could address the comment regarding best place to contact? A dedicated Lemmy community perhaps, spread the fediverse love?

I understand your hyperfocus for the app, because I assume most researchers or PhD candidates are also hyperfocused. I figured that news releases regarding the field of research is also relevant in some way, finger on the pulse sort of thing. Probably not as easy as grabbing the papers as you currently are.

Thanks! That was an insightful read. Didn't know models/learning could be this efficient and small. Main reasons why I asked is, I saw another app on Reddit that mentioned learning is done locally, but others found out it was alloted to gemini on users' phone (which the dev insisted was done locally but not really privacy respecting imo) and almost every app with such a feature was humongous. Thanks for making Aftergleam and keeping it open source

That's what a static forum with comments is for 😂

Using discord/matrix for this means that to get to a recipe, you have to scroll through 100 comments.about recipes not linked to the recipes, and you can't browse the recipes. You just have to be there or never know it happened pretty much.

Also why support "found" using discord like zephyr have almost 0 help on them at all.

I also like Shikantaza :)

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

https://github.com/jakobkreft/Aftergleam/issues/3

I can see that the crux of my problems drilling down on environmental topics may be made more difficult by some missing sources as illustrated in this issue. I'm still training away and getting some good hits so it's working regardless.

Your app has a lot of potential. I'm going to use it as a positive/factual news source for a while.

Beacuse the first folder in the git is ".claude/skills/spec".

Hi, I've been using the app ever since it was on F-droid. I wanted to ask you about the learning/model. The app is under 2MB, so where is the model/learning done ? Is the learning assigned to gemini ?

This is fantastic, I've been curious for a way to more easily consume science papers than finding them in the wild. This answers that curiosity perfectly! I look forward to the education, thank you

Because it shows it’s LLM made.

Examples that are interesting to me by selecting 2 initial biology and ecology topics and then adding keywords ("soil" and "biochar"):

Coarse Woody Debris of Mixed Tree Species Overrides Soil Matrix Constraints on Microbial Richness and Abundance in an Ancient Broad-Leaved Forest https://www.biorxiv.org/content/10.64898/2026.10.09.757649

Found with Aftergleam, a privacy reader for arXiv scientific papers.

Synergistic effects of biochar and vegetation on microplastic and nanoplastic removal from stormwater runoff in bioretention systems https://doi.org/10.26434/chemrxiv.15010823/v1

Found with Aftergleam, a privacy reader for arXiv scientific papers.

Like I said, I'm not a scientist so I'm using keywords to hit terms that are relevant to me like "tree", "species", "biochar", "plant", "soil" etc. Then I deleted some after training and added "Australia" "forest" to train more relevant to my location to see what would happen. Then I might add my climate type later to drill further down.

So I can't really search to find specific papers as I don't technically know what I'm looking for. I'm extrapolating papers that might be relevant to my environmental line of work. Edit: Searched some individual keywords and I understand what you mean, it's just time to select the papers that are relevant in that particular keyword. I'll give that a try for a while and keep my keywords to broaden the feed.

If you were going to develop the app further, it could become quite a nice mixed feed reader for specific topics of interest, maybe with a sharing intent for social media too. I could imagine a News tab separate to For You and Explore that picked up scientific news releases based on keywords as well coming from the main journalistic sites like phys.org, theconversation, mongabay etc.

Thanks for your help.

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!

I've installed it and give it a try. At the first look, it seems very Nice !! Thanks

With further keyword tweaking, I can get more relevant papers. As the keywords have a limit, if I train it with a certain list of keywords, then delete them and add new ones for further training, is this a way around that limit? Will the training stick for deleted keywords?

I don't know enough but I feel as though this app will be very helpful for the people that require it. Well done.

Amazing! Thank you

I'm not a scientist, nor do I have a "field" but I do like what I like and it managed to find somewhat relevant papers in one go (it complained about it though) and that's without further tweaking in the settings.

Very nice. Clean app too.

Here's the F-droid link

https://f-droid.org/packages/si.jakobkreft.aftergleam/