Pinned Tweet
Picked up Inference Engineering because I think it's a powerful intersection of the things I enjoy and it's going to be in any and everything. I'd keep sharing the things I'm learning on this thread.
As a scalper, I want to trade the setup I’m looking at without opening a sidebar to enter my position size.
That’s what we’ve built on Furge.
Add a long or short drawing, drag it to adjust your risk, and size your position directly on the chart.
The small popover gives you the controls: amount, leverage and which exchange to trade on.
Hit “Trade” and the order goes through. No extra confirmation step.
From setup to execution, right on the chart.
6/n
The KV(key-value) cache saves the model from recalculating information about previous tokens every time it generates a new one.
That saves work, but the cache takes GPU memory. Longer conversations mean more memory per request.
Adédọ̀tun𓂀 retweeted
I have two goals for exercising, to stay fit and not have potbelly. All these biceps, triceps, traps etc no concern me.
C43 AMG: Stage 1 + downpipe + full valved catback 🔊 (built ts btw 😝)
youtube.com/watch?v=i9w-ws85…
Adédọ̀tun𓂀 retweeted
Solo in Kenya - The Adventure Of A Lifetime, Out now on YouTube! The past one month has been surreal to me, I am so happy I am getting to make things like this and share it with y’all, thank you for supporting 🥹🙏🏾✨
Adédọ̀tun𓂀 retweeted
Introducing Model Manifest, MoM for short.
mom.yaml sits in your repo and decides which model handles each message.
Small stuff goes to the cheap kid, hard stuff goes to the smart one, and mom doesn't wait for the cheap kid to fail first
I ran HumanEval on Claude Opus 5 alone: it solved 163 of 164 problems and the run cost $1.34. Then I ran the same problems through a 15-line mom.yaml.
Every problem started on Qwen3 Coder Next. Eight hard ones escalated to Sonnet, and one of those went up to Opus. It solved all 164, and the run cost 9 cents.
The file beat the best single model and cost 93% less, because 156 problems never needed an expensive model.
MoM is an open protocol, not an Aster feature. Any tool can read the file. Aster is the reference implementation.
Adédọ̀tun𓂀 retweeted
Once you’ve successfully made money through a repeatable, non-random process, making money again often becomes easier because now you have both capital and a mental model for how money is made. It’s actually a game
Adédọ̀tun𓂀 retweeted
Recently wrapped up my undergraduate degree, finishing with first-class honors.
Here's a recap of my journey:
– Software engineer intern at @Google (twice)
– Software engineer intern at @Bloomberg
– Software engineer intern at Fermyon (now part of Akamai)
– Software engineer intern at Indicina
– Spent 2 years at Terrace building an aggregated crypto liquidity platform
– Spent over a year at BuildQL building infrastructure that helps people learn to build software
– Co-hosted @TechfiestaHQ, celebrating and recognizing top talent in the industry
– Spoke at several tech events.
– Served as technical lead for GDSC LAUTECH
Now, I'm building @getdriftless to keep your database always in sync with Stripe.
If you're at Google in EMEA and your team is looking for an L3 (I'm already in the conversion pipeline), please reach out!
Learn more about Quyum, a returning software engineering intern based in Zurich! Explore our open roles and opportunities ➡️ goo.gle/4dG74eB

ALT Why did you apply to Google? Quyum: “Working at Google has been a dream of mine since I started learning how to code in high school. I knew I wanted to work at a place where I could build products used by billions of people. I also had a great experience as a Google intern in past summers, so that inspired me to return.”
Adédọ̀tun𓂀 retweeted
The CfP for SysConf 2026 is now open!
If you have tackled interesting technical challenges (successfully or otherwise), are researching a frontier area or spent some time in the weeds of a technical area, we’d love to hear from you: sessionize.com/sysconf-2026. Our CfP is open until 11:59 pm (WAT) on August 22nd, 2026.
























