Umur Cubukcu
@umurc
Followers
2K
Following
2K
Media
36
Statuses
723
Founder at Ubicloud. Previously visiting partner @YCombinator, founder @citusdata (acq by @Microsoft), led PostgreSQL product @Azure
San Francisco
Joined April 2009
More agents running on @UbicloudHQ -- and managing their memory via Ubicloud PostgreSQL. Our customers @voltagent_dev make that super easy.
🚀 Managed Memory is live! No infra. No schema headaches. Spin up a hosted memory DB in minutes & plug it straight into your agents. 🌍 US + EU regions ⚡ Conversation + working memory 🧠Optional vector search runs on @UbicloudHQ 🔥
1
4
9
Not sure but it might be a short facebook post by @nixcraft that snowballed into it. 🙏 Wonderful to see it resonate, thanks for the share https://t.co/22HtEO7xlA
github.com
Open source alternative to AWS. Elastic compute, block storage (non replicated), firewall and load balancer, managed Postgres, K8s, AI inference, and IAM services. - ubicloud/ubicloud
0
0
7
I really enjoyed speaking with @oguzhan at his CNBC-e show last week. The weather really cooperated, too, as we talked about entrepreneurship, open source, and of course, AI. // Oğuzhan ile çok keyifli bir sohbet oldu CNBC-e Vadi Buluşmaları'nın son bölümünde. Kameralarla
1
0
15
If you're ever curious about how inference requests work under the hood, you'll find this walk-through informative -- all the way down to microseconds.
Ever wondered how LLMs serve millions of requests? We traced every microsecond of it. https://t.co/ZPusIpcHD8 vLLM is the most popular open source inference engine. We couldn't find a reference document that describes the lifetime of an inference request. So, we put it
0
2
12
Cloud infra --in this case, managed Github Actions runners-- that is not just 80% lower in cost, but also 3x faster ⚡️
s/o to @UbicloudHQ Helped speed up our CI tremendously. With some caching from ~8 mins to now only ~2.5 mins Having that faster feedback loop is amazing
1
0
14
How we implemented VMs with shared CPUs that burst 2x under load (now trending on HN) ⚡️ Result: Highly performant cloud instances with dedicated NVMe storage, starting at just $0.01 per hour
We implemented burstable instances using Control Groups v2 (cgroups v2), a Linux kernel feature that helps manage resource usage. As an open source cloud, we’re committed to transparently sharing everything we do. We had four key takeaways and talked about them in our blog post.
0
1
14
We've built a deep research agent, using only open source models. It ended up a 400-line Python program, performing remarkably well vs. proprietary solutions (without the bells and whistles, of course). Step-by-step instructions to replicate it, and our lessons learned while
0
1
17
It took less than 20 years for the former to result in the latter in Turkiye. If unchecked, I worry about it taking shorter in the US.
The line between “if you are vocal in your opposition to the president’s policies, I shall jail you” (Mahmoud Khalil, student in the US) and “if you are the leader of the political opposition to the president, I shall jail you” (Ekrem Imamoglu, major of Istanbul) is dangerously
0
0
7
The line between “if you are vocal in your opposition to the president’s policies, I shall jail you” (Mahmoud Khalil, student in the US) and “if you are the leader of the political opposition to the president, I shall jail you” (Ekrem Imamoglu, major of Istanbul) is dangerously
1
0
11
Y Combinator has helped create more than $800 billion in market value over 20 years. Thank you to Paul, Jessica, Trevor and Robert and Happy Birthday to the fixed point combinator that changed the world.
0
0
13
Our co-founder Ozgun speaking at #WebSummitQatar on reasoning models and @deepseek_ai. Happy to deliver the best of open source AI models to wherever you are #opensource #Ubicloud
0
4
7
We love orange at @UbicloudHQ :) (and @ycombinator , too)
Every aspect of life is being stripped of color. Many have noticed this trend — but why exactly is it happening? Something deeper is going on… (thread) 🧵
0
1
7
Among many of his insights, this is one of my favorites. https://t.co/4VD1Oe3n7r When you do something you love, for a long time, and with care -- things compound.
1
1
10
Happy customers running on both x86 and @Arm on @UbicloudHQ
@UbicloudHQ has the best GitHub runner experience. Thanks to @menescakir for chasing down a GitHub issue and helping us with https://t.co/Tq6wEIS41p. Cut our build time from 15+ minutes to under 4 minutes. Oh and all this while being 10x cheaper than the official GitHub runners!
0
0
6
New 3h31m video on YouTube: "Deep Dive into LLMs like ChatGPT" This is a general audience deep dive into the Large Language Model (LLM) AI technology that powers ChatGPT and related products. It is covers the full training stack of how the models are developed, along with mental
779
3K
20K
