Ryanne Dolan
@ry.codes
BigTech BigData infra @ LinkedIn, Twitter, Cloudera
In the early days of LinkedIn there was a period of "inversion", which was what they called it when "your tech debt interest is greater than your tech production" (your quote), so they stopped all new feature work and rewrote everything. Took years!
November 10, 2025 at 9:52 PM
In the early days of LinkedIn there was a period of "inversion", which was what they called it when "your tech debt interest is greater than your tech production" (your quote), so they stopped all new feature work and rewrote everything. Took years!
Reposted by Ryanne Dolan
Need a Sandwich Artist
November 4, 2025 at 5:46 PM
Need a Sandwich Artist
A one-line horror story just in time for Halloween!
October 31, 2025 at 7:29 PM
A one-line horror story just in time for Halloween!
Oh no I placed a few bids 😶🌫️
October 24, 2025 at 1:45 AM
Oh no I placed a few bids 😶🌫️
He was great in The Hurt Locker.
October 22, 2025 at 2:20 PM
He was great in The Hurt Locker.
My team has done it the other way around: we started with ETL and CDC and accidentally built a federated query engine.
October 18, 2025 at 2:23 PM
My team has done it the other way around: we started with ETL and CDC and accidentally built a federated query engine.
I guess a big reason for using tokens is that you can easily snapshot and restore state. Rather than talking to a single machine with gigabytes of state, you can load a few kilobytes into any machine with the model and resume where a previous machine left off.
October 12, 2025 at 3:33 AM
I guess a big reason for using tokens is that you can easily snapshot and restore state. Rather than talking to a single machine with gigabytes of state, you can load a few kilobytes into any machine with the model and resume where a previous machine left off.
The chair moves with her, as if she's sitting on it. Who knows.
October 11, 2025 at 2:43 PM
The chair moves with her, as if she's sitting on it. Who knows.
Yes but loop+tools can solve problems that are fundamentally not one-shot-able. It's hard to appreciate this aspect of LLMs until you give them a complex tool to play with.
September 20, 2025 at 2:38 AM
Yes but loop+tools can solve problems that are fundamentally not one-shot-able. It's hard to appreciate this aspect of LLMs until you give them a complex tool to play with.