Except as the corporate spending card company they see just how much their customers are spending on inference and have made it a cornerstone of their marketing for the last two years.
Their positioning is: You already have all your teams spending routing through us so that through greater transparency to finance you get better spending controls. We now apply the same to inference, which is spread out across all parts of your company in a similar way.
All summed up really well with their tagline “Tokens are money. Save both.”
Don’t think I’ll be using model routing / shared gateway from Ramp any time soon, but the value of this kind of capability is definitely there. And there’s no shortage of vendors aiming to offer this kind of gateway/router functionality. But most of those come from the technology side of the organization. This gets to come from the office of the CFO.
I was about to link to the same "What's on the Menu" site, which I remembered being an impressive site for library digitization project years ago, but it was apparently retired in January 2025.
Now [1] redirects to [2], essentially an About page with links to the data.
To be fair we were done transcribing and double verifying in about 2015. We’d get a batch of new menus from the imaging team, tweet it and no matter the size it’d be gone in about an hour.
This absolutely rocks: like an open and usable version of the once beloved and now deprecated Echo Nest Audio Analysis api [0]. There used to be a tool to use this data to manage your own (Spotify-only) music library and while it wasn’t magic, it sure felt like it.
11 years ago, NY Magazine published my favorite startup zeitgeist article of the Uber-For-X era: “Let’s, like, demolish Laundry”. This feels like a perfect revisit of that for our darker and more cynical era.
There have been several projects over the past few years to make text embeddings visually explorable (notably Nomic.ai's Nomic Atlas). However, Apple's just released a tool that makes this kind of analysis super accessible and insanely interactive.
Under the hood it's powered by Mosaic[0], a dataviz library built on top of DuckDB that's designed to handle coordinated interactive plots over huge datasets, the kind of thing where you interact with one plot and the rest all respond, which requires going back to the database to recalculate all the aggregations.
I've been fanboying Mosaic for the past year but finally have this to point to as an illustration of what's possible with it.
> Polygons, curves, or lines are an afterthought. For example, the polygonToCells function returns a set of cells that are entirely within the polygon. If you want to join, you'd need to also have the set of all cells that entirely contain the polygon. I've never found a reliable way to get that.
With v4 of h3 they (finally) have a clean syntax for this with polygonToCellsExperimental[0].
Now there’s options for
- Cell center is contained in the shape (default)
- Cell is fully contained in the shape
- Overlapping (covering): Cell overlaps the shape at any point
- BBOX: Cell bounding box overlaps shape
Makes life a fair bit easier if you’ve gotta deal with H3 polys. And if you’re working locally, DuckDB Spatial’s r-tree indexing[1] can make for a nice stand-in for PostGIS as a quick point-in-polygon solution without the need to spin up a service.
Nice, I did not know about polygonToCellsExperimental.
As for DuckDB & rtree, it's alas not a replacement of postgis yet and the indices cannot be used (yet) in joins. In fact, I even have workflows where I iterate over rows in python and run duckdb queries one after the other rather than joining in just one query because of this very issue.
Chronicling America is great, but as you’ve noticed it’s limited to issues that are in the public domain. It’s not just LC, the National Digital Newspaper Program[1] has been funding hundreds of libraries across all 50 USA states to make it possible since the mid-00s.
Australia’s national library offers Trove[2], which has a huge collection of Australian public domain newspapers.
Most of their repository is likely funded by your tax dollars and is there for the public to use.
Trove is fantastic. It came in handy very recently for the same project I used the Library of Congress for[1]. The Museums of History New South Wales has a great online archive as well[2].
Their positioning is: You already have all your teams spending routing through us so that through greater transparency to finance you get better spending controls. We now apply the same to inference, which is spread out across all parts of your company in a similar way.
All summed up really well with their tagline “Tokens are money. Save both.”
Don’t think I’ll be using model routing / shared gateway from Ramp any time soon, but the value of this kind of capability is definitely there. And there’s no shortage of vendors aiming to offer this kind of gateway/router functionality. But most of those come from the technology side of the organization. This gets to come from the office of the CFO.
reply