HackerTrans
TopNewTrendsCommentsPastAskShowJobs

dhrm1k

no profile record

Submissions

Show HN: ArchiveJump – Browser extension to instantly jump to archived articles

github.com
3 points·by dhrm1k·w zeszłym roku·0 comments

Depict: AI-powered search for Indian museum artworks using CLIP embeddings

github.com
1 points·by dhrm1k·w zeszłym roku·1 comments

comments

dhrm1k
·29 dni temu·discuss
are there any related to networking? i contribute to a lot of openwrt packages (well mainly, only the ones i use and would love to find a few fun tools around that domain.)
dhrm1k
·3 miesiące temu·discuss
ah nice. i’ve actually been building something pretty similar on top of wireguard too

my use case is a bit different though. i started because i wanted to give friends access to specific things in my homelab, but very selectively. like “you can use jellyfin on this one machine, but you can’t ssh, and you can’t even see my other devices”

tailscale is honestly amazing for getting devices connected, i still use it a lot. but once i started trying to do these very specific “this machine can talk to that machine only on this port” kind of setups, it started feeling more complex than it should be, at least for personal use. ACL editor is more confusing when it comes to this. i know we have got option for tags and things, but those are very poorly documented and i haven't found a single tutorial that works nicely.

your userspace approach is really interesting btw, especially the no tun / no root part. makes sense to run it on rigit enterprise environments.
dhrm1k
·11 miesięcy temu·discuss
pls add a readme to the repo.
dhrm1k
·w zeszłym roku·discuss
wait, how did you get irc dot com?
dhrm1k
·w zeszłym roku·discuss
Hi HN!

We built Depict (https://github.com/dhrm1k/depict) - an AI-powered search engine that lets you find Indian museum artworks using natural language descriptions like "painting of fruit" or "blue sculpture."

The project uses OpenAI's CLIP model to understand both text and images, creating embeddings that enable semantic visual search across collections from major Indian museums (National Gallery, National Museum Delhi, Indian Museum Kolkata, etc.).

Key technical details: - Scraped data from Museums of India government portal - Generated CLIP embeddings for artwork images - Built React frontend + Python backend on Modal AI - Similarity search using cosine distance in 512-dimensional space

The live demo currently includes ~2 museums due to compute costs, but the full pipeline is built to handle all 6 major collections. It was inspired by @ekzhang's dispict but focused specifically on making Indian cultural heritage more accessible.

Would love feedback on the search experience and any ideas for improving discoverability of museum collections!

Tech stack: Python, React, CLIP (ViT-B/32), Modal AI, h5py Repo: https://github.com/dhrm1k/depict
dhrm1k
·w zeszłym roku·discuss
[dead]
dhrm1k
·w zeszłym roku·discuss
funny I got to know about ersei from Purdue linux group's members list and here he is on the frontpage of hn.
dhrm1k
·w zeszłym roku·discuss
i mean what's the breaking point? why is ublock no longer in the store whole the lite is?
dhrm1k
·w zeszłym roku·discuss
it indeed is. when I saw it in my mstdn feed, i too was shocked but upon reading, uh, no.
dhrm1k
·w zeszłym roku·discuss
[dead]