SMLL: Using 200MB of Neural Network to Save 400 Bytes
February 6, 2026
We compressed Jane Austen to 10 bytes. The model weights required to decompress it are 200 megabytes.
Hi, I'm Frank! I write code that keeps you safe and private online.
I'm a software engineer passionate about privacy, security, embedded systems, and open source. I work mostly with Go, TypeScript, Python, Java, Kotlin, Swift, and I'm currently learning Rust. I recently started operating my own ASN AS402030 to learn more about BGP and peering.
I am currently attending the Summer 2 Fall 2 (I extended 🙈) batch of Recurse Center, having recently left my role as tech lead and engineering manager at JPMorgan Chase's Global Business Accelerator. Previously, I was a founding engineer at Svix (YC W21) and lead backend/desktop engineer at Ghostery/Cliqz, working on anti-tracking software.
Probably doing something I shouldn't with cGo or WinAPI.
February 6, 2026
We compressed Jane Austen to 10 bytes. The model weights required to decompress it are 200 megabytes.