We consume a lot of content today — books, articles, substack, youtube videos, netflix, audiobooks - and forget most of it.
We don't really notice how much we've consumed, or
what shaped us. Goodreads could have been the place to track all
of this, but it stopped at books and stopped evolving.
TNG is what Goodreads could have been: a log for every kind of
media.
designed to be private, fast, and pretty and very useful if you're a polyglot
Nothing AI about it *yet* More old school - totally offline if you want, free, no social features if you don't want them, fast.
What
Privacy-first by default. Supports all media types you can think of - books,
articles, newsletters, papers, podcasts, web novels, streaming services - and treats polyglot reading as a
first-class case rather than an afterthought (because after all I built this for myself mainly)
Sometimes you're reading a book and listening to a book at the same time. You can track them separately or track them together. Sometimes you're reading a different edition of the book. You can track that separately as well. Sometimes you've watched a movie based on the book. Now you can track that as well. Sometimes you're trying to figure out what you should read next or what is the next book in the sequel. Easy tap.
Unique opt-in social features that don't reduce to a follower graph. Ability to track reading groups.
The plan is to eventually layer in AI for genuinely useful recommendations grounded in your real consumption history, rather than a platform's idea of what you should engage with next.
A quick prototype I mocked up for a job interview, based on
the conversation with the interviewer. The point was to show
how fast you can stand up something AI-driven that delivers
real value end-to-end.
What
A tool that triages incoming requests from sales - sorting and
ordering them so software and product teams can roadmap against
them without the usual manual back-and-forth. The interesting
bit is less the model and more the shape of the workflow: turning
a noisy inbound stream into something a planning meeting can
actually use.
I'm learning Finnish; a friend is learning Estonian. Both
languages have very few good tools for learners. We were
inspired by JPDB.io
and decided to build our own - partly to have full control over
the experience, partly to learn how parsing actually works. If we own the parser, we can improve it. On the roadmap is making the parser improve automatically *hint* - karpathys autoresearch + user input.
What
FinnEst is a continuously-updated, state-of-the-art parser
for Finnish and Estonian, wrapped in a science-based language
learning app. Paste in any text; it tracks the words you
already know, builds review decks from what's new, and uses
spaced repetition to lock things in. Language learning is
a passion, so this one is built to scratch a real itch.
ICEYE operates the world's largest commercial SAR satellite
constellation.
The interfaces through which customers task,
monitor, and retrieve data from that constellation are the
product.
Getting them right is critical - because to the end user the interface *is* the constellation.
For governments, insurers, and defense customers alike, these interfaces turn a fleet of satellites into something as easy to use as ordering someting from Amazon.
What
I led the product for both the API and the UI — the two
primary surfaces through which the world interacts with the
constellation. A patent application is pending on the work.