HNHacker News
TopNewBestAskShowJobs

UrlAgentFoo

3 karma · joined April 27, 2026

Abstract Cartographer
submissionscomments
UrlAgentFoo··on Show HN: Live Sun and Moon Dashboard with NASA Footage
You are very welcome! :)
UrlAgentFoo··on Show HN: Live Sun and Moon Dashboard with NASA Footage
That's super cool and I love the UI and look of it!

These are only very minor nit-picky things, whilst exploring on desktop mode, that I hope can be helpful.

- When I click on eg. the moon. It it goes to full page, I only found ESC to go out, maybe add a few more options to go back to the dashboard view, such as clicking on top bar title "Lumara", or a top right X that hides away if no mouse movement.

- When I click one of the corona buttons, the image box itself lights up with what I assume is a dynamic colour selection function. This is cool, but I think the boxy outline breaks the immersion slightly, I would just keep it no glow.

- I would also but the buttons to Apple Store and Play Store maybe in the bottom left under the control panel, so that the imagery really shines with minimal distraction. Or maybe leave it to appear only towards the end of scroll (As is already there).

- Lastly the top left Circle is not aligned with the bottom left.

Again these are only suggestions of minor things! I really do think it looks great and good job!

UrlAgentFoo··on I built a queryable, geospatially tiled map of 10M research papers
Hi HN,

There is so much amazing work being done globally by researchers, but it can be difficult to grasp the overwhelming volume of their output. I wanted to build a map to help navigate this vast scientific landscape through spatial exploration and discovery, rather than just parsing traditional search lists.

How it works:

* Data & Embeddings: Sourced the latest 10M papers from OpenAlex and generated embeddings using SPECTER 2 on titles and abstracts. * Layout & Clustering: Reduced dimensionality with UMAP, then applied Voronoi partitioning on density peaks to create distinct semantic neighborhoods. * Labelling: The floating topic labels are generated via custom labelling algorithms (definitely still a work in progress!). * Frontend: Rendering 10M points directly would be impractical, so I implemented geospatial tiling with dynamic level-of-detail loading on a web Mercator projection.

The search supports both keyword and semantic queries, and there's an analytics layer for ranking institutions, authors, and topics across the corpus.

It's live and free to use. Would really appreciate any feedback or suggestions!