About
I was born in Istanbul in 1995. After two years of engineering study adventure in CA, I…
Activity
1K followers
-
Enes Çakır reposted thisEnes Çakır reposted thisWe've been building managed Postgres for 15 years. Across all of them, one configuration remained constant: strict memory overcommit. Then, a three-character Linux kernel bug hit us. We had to temporarily disable this setting and figure out why our customers' databases didn't have enough memory. Attached pic shows the fix from Linus. Kudos to Burak Yücesoy for driving this investigation and rolling out the new kernel version in our fleet. In the process, Burak also wrote down all of this together in a nice blog post. - Linux out of memory (OOM) killer - Why PostgreSQL can't tolerate the OOM killer - How Ubicloud solves this issue: fail early, not catastrophically - The three-character Linux kernel bug - How we set the linux commit limits - Our open source implementation One of the most interesting blog posts I've read in a while. Link is in the comments. Please feel free to chime in with any questions and comments. ⬇️⬇️
-
Enes Çakır reposted thisEnes Çakır reposted thisI am excited to announce our partnership with ClickHouse. And it’s much more than a logo swap. What this means in practice: ClickHouse is launching managed Postgres, powered by Ubicloud. It is fast, tightly integrated with ClickHouse for analytics; with native capabilities to sync and query data across datastores. What makes this different is how we’re doing it. This isn’t a thin integration. The ClickHouse team is building directly with us and contributing to the Ubicloud open source project with shared code, shared goals, and daily engineering collaboration. There’s also history here. Our teams built and operated Postgres before at Microsoft, Citus Data, and earlier Heroku Postgres. When ClickHouse looked for a Postgres partner, they needed a team that had already done this multiple times before. That’s what Ubicloud brings. If you’re building systems that need fast transactions + real-time analytics (including AI workloads), this removes a lot of complexity. Thrilled about what's ahead. Blog post is live - link in the comments.
-
Enes Çakır reposted thisEnes Çakır reposted thisPGConf.EU rotates European capitals, but always feels like home! This year, we're excited to announce Ubicloud Postgres on AWS. Compared to AWS Aurora: * Aligned with the ethos of Postgres, everything in our managed db (control plane, etc.) is open source * 3-6x higher performance for IO bound workloads in TPC-C / TPC-H * Lowers cost to slightly over half to a third of the cost in the same benchmarks, on AWS If you're at PGConf.EU, please swing by our booth and tell us what you think! Link to our launch is in the comments. PS: Thank you PGConf.EU organizers for making this conference feel like home every year.
-
Enes Çakır shared thisNext stop: 10k stars! 🚀Enes Çakır shared thisWe had our team retreat last week and our repo's star count jumped to 9.7K on GitHub. Coincidence? Yes. First, a user asked Grok for an open source alternative to AWS. Then, nixCraft (Linux Tips, Hacks, Tutorials) shared Ubicloud on Facebook. This put us on the list for GitHub trending repos - first for Ruby and then globally across all languages.
-
Enes Çakır reposted thisEnes Çakır reposted thisExcited that Ubicloud is the top trending repo on GitHub! Who would have thought that Linux Facebook groups carry so much weight these days? If you’re into open source, please share the love on Ubicloud’s GitHub repository. ❤️
-
Enes Çakır shared thisWe offer the best price-performance on the market, even for high-performance runners 🚀Enes Çakır shared thisI'm excited about our new premium runners for GitHub Actions! The biggest pushback we received so far is that the feature "sounded too good to be true." So, Enes Çakır wrote a blog post that shares benchmark results and quotes customers who started using premium runners in the past two weeks. Four highlights: 1. We use gaming CPUs to improve your CI/CD performance by another 40%. At 1/5th the price of GitHub hosted runners, you get 10x better price-performance 2. 10x more free cache space 3. Secure: New and ephemeral VMs for each job, encrypted at rest and in-transit, with SOC2 compliance 4. Seamless upgrades. You can literally upgrade to premium runners or downgrade back to standard ones by clicking a toggle button If you have questions or you'd like credits to try out premium runners, please drop us a line at support@ubicloud.com! https://lnkd.in/e6py88BKUbicloud Premium Runners: 40% Faster Builds, 10x Larger CacheUbicloud Premium Runners: 40% Faster Builds, 10x Larger Cache
-
Enes Çakır shared thisOpen source, blazing fast, and built by folks who actually know Postgres. Ubicloud PostgreSQL v2 is hereEnes Çakır shared thisToday, we're excited to announce Ubicloud PostgreSQL v2: Seven new features, higher performance! https://lnkd.in/eTDxw4bc How is this different than other managed Postgres solutions? 1. We use local NVMes. This helps us deliver price-performance that's 7.9 - 11.1x better than Aurora Postgres. 2. Our team has years of experience with Postgres. This makes us keep reliability as a forethought. 3. Ubicloud provides the first open-source solution for managed Postgres. This keeps you in control of your database. If interested, check out our v2 and tell us what you think! ⬇️⬇️⬇️Ubicloud PostgreSQL: New Features, Higher PerformanceUbicloud PostgreSQL: New Features, Higher Performance
-
Enes Çakır reposted thisEnes Çakır reposted thisToday, we're excited to announce Ubicloud PostgreSQL v2: Seven new features, higher performance! https://lnkd.in/eTDxw4bc How is this different than other managed Postgres solutions? 1. We use local NVMes. This helps us deliver price-performance that's 7.9 - 11.1x better than Aurora Postgres. 2. Our team has years of experience with Postgres. This makes us keep reliability as a forethought. 3. Ubicloud provides the first open-source solution for managed Postgres. This keeps you in control of your database. If interested, check out our v2 and tell us what you think! ⬇️⬇️⬇️Ubicloud PostgreSQL: New Features, Higher PerformanceUbicloud PostgreSQL: New Features, Higher Performance
-
Enes Çakır reposted thisEnes Çakır reposted thisAbout a year ago, Hetzner launched their leading bare metal line, AX 162. At Ubicloud, we were very excited to adopt it, but soon encountered reliability issues. We observed that the new servers had 16x higher annual failure rates than their predecessors, the AX 161 line. After months of debugging and working with Hetzner, the solution came after several hardware updates. Although the journey was painful, we learned a lot from it and wanted to share our experience. https://lnkd.in/ecANTz6f Three insights: 1/ You can reason a lot about the underlying hardware by using Linux tools such as powerstat, sensors, lshw, and dmidecode. 2/ The AX 162 offers great price / performance, so we quick to adopt it. We should have done a more gradual roll out, starting with non-critical workloads. 3/ Ubicloud should have been running on more bare metal providers. Today, we already support two more providers and we’re working on adding the fourth one. Today, we’re happily running on Hetzner, Leaseweb, and Latitude.sh. If you’re looking to learn more about running on bare metal, we hope our blog post serves as an interesting resource!Debugging Hetzner: Uncovering failures with powerstat, sensors, and dmidecodeDebugging Hetzner: Uncovering failures with powerstat, sensors, and dmidecode
Enes Çakır commented on a post
1y
🎉 Excited for you
-
Enes Çakır liked thisEnes Çakır liked thisAWS customers are reporting cloud bills worth billions of dollars. Come to us, we're cheaper! 😀 (^Ubicloud)
-
Enes Çakır liked thisEnes Çakır liked thisFrontier-level AI safety research doesn't only come from Big AI labs. It comes from our ecosystem too. Ubicloud just published a novel AI safety study. The study evaluates AI safety properties of popular open-weight models, namely DeepSeek V4, Kimi K2.6, and Nemotron 3 Ultra. It's novel research highlighting some important new findings based on: → Five safety dimensions, from jailbreak robustness to data protection → Standardized benchmarks plus adaptive, autonomous red-teaming → A finding the big labs' benchmarks missed: models that refuse in their visible answer while leaking secrets in their reasoning trace On standard benchmarks, open-weight models look as safe as the frontier. Under adaptive attack, they drop 35 to 44 points while the frontier baseline barely moves. Benchmark parity does not deployment parity, an important finding now founders are looking for frontier alternatives The study demonstrates that our ecosystem possesses the deep expertise required for frontier-level AI work, not just building on top of existing solutions, but thoroughly interrogating them, executing rigorous red-teaming, and publishing valuable insights for the benefit of the entire industry. Every founder building with open-weight models now has a sharper playbook: add a purpose-built guardrail layer, treat the reasoning trace as real output, and stop reading benchmark scores as safety guarantees. That's ecosystem depth. Companies doing the hard research, then giving it away so everyone builds better products. Congrats to the Ubicloud and Superagent (YC W24) teams, for setting the bar. Full study: https://lnkd.in/ef_Gsk_e #DutchTech #AISafety #EcosystemAI Safety Study for DeepSeek V4, Kimi K2.6, and Nemotron 3 UltraAI Safety Study for DeepSeek V4, Kimi K2.6, and Nemotron 3 Ultra
-
Enes Çakır liked thisEnes Çakır liked thisAI safety studies for open models barely exist. We’re excited to publish the first study that evaluates DeepSeek AI V4, Kimi (Moonshot AI) K2.6, and Nemotron 3! Here are three key findings: 1/ On the benchmarks everyone knows, open-weight models keep pace with Claude Opus 4.8 (our frontier baseline). Kimi and Nemotron score within 1.4 points of Opus 4.8. A user who stopped here would reasonably conclude that open-weight models are about as safe as the frontier model. 2/ On adaptive, autonomous red-teaming, the picture changes. Opus 4.8’s safety scores mostly hold (-5). Kimi K2.6 (-36), Nemotron 3 Ultra (-43), and DeepSeek V4 Pro (-44) all see significant drops. 3/ Across the five safety dimensions we evaluated, data protection is weak across the board for open-weight models. Prompt injection is Nemotron's worst dimension; trace-leak jailbreaks are DeepSeek's. If you’re deploying these models, spend effort where your AI model is weak. Also excited to have collaborated with our fellow YC batchmate Superagent (YC W24) on this. Link to the study is below! If you have any questions or feedback, please drop us a comment below.
-
Enes Çakır reacted on thisEnes Çakır reacted on thisFeel privileged to be working with Jeremy Evans! Just saw that he has Messi level stats on Gitfut. 😀 Also what better way to celebrate open source maintainers than turning their work into Wold Cup style player cards. * Generational talent: hall-of-fame maintainer: high and balanced, earned over years. * Work rate: High / High * Style: Explosive This one goes out to Jeremy and all open source maintainers who make our tech stacks better, bit by bit.
-
Enes Çakır liked thisEnes Çakır liked thisLast week, we migrated a customer to Ubicloud PostgreSQL. Shortly after the migration, they reported a 30% improvement in one of their heaviest API endpoints. It was a pleasant surprise for them, but it is just another Tuesday for Ubicloud. By using local NVMe SSDs instead of network-attached storage, we can deliver significantly lower latency and higher throughput than traditional managed PostgreSQL offerings. But the story got more interesting. Two days later, they shared new numbers showing 2-3x faster execution times for several of their most frequently executed queries. And yesterday, they sent another update. A few queries that were previously among their top 50 by total database time spend dropped from 1.7 milliseconds to just 50 microseconds. That's a 34× improvement. I hesitated to share these numbers because they almost sound too good to be true. But this is what happens when you combine local NVMe storage with over 15 years of experience building and operating managed PostgreSQL services. PostgreSQL performance is often discussed in terms of query tuning and schema design. Those matter. But the underlying infrastructure matters too and sometimes a lot more than people expect.
-
Enes Çakır liked thisEnes Çakır liked thisKimi is the only AI model that comes with a tool to verify correctness. We tried it on vLLM and were surprised by the results. On tool calling schema accuracy, we saw 91% pass rate. That means one of ten tool calls had issues in generating the correct tool call syntax. That doesn't sound so bad, but if you have a workflow that makes 10 tool calls, your results are only 39% likely to have all tool calls succeed. This is one of the reasons why people think open weight AI models don't perform as well as their benchmarks. Depending on the inference engine or provider you use, your AI model doesn't deliver its capabilities. And this isn't an AI model problem. It's a system problem. We need classical software engineering to guard the model to only output valid tool calls. At Ubicloud, we've been carrying a vLLM patch (in our open source fork) that uses constrained generation to fix this issue. Fortunately, the vLLM community recently accepted a patch that notably increases tool call success. The changes should go out with the next release. If you're interested in this space, please drop a note below. We may then blog about this topic. 👇👇
-
Enes Çakır liked thisEnes Çakır liked thisThe GPU shortage just became the CPU shortage. News from Hetzner and Microsoft dropped yesterday; and they are wild. First, Hetzner, the world's leading independent bare metal provider. A few months after raising prices by 30%, Hetzner hiked up prices by another 3-4x. The new prices are 407% to 424% of what they were just two months ago. AX102 Series: €107/mo -> €454/mo AX162 Series: €207/mo -> €844/mo For context, **when Russia invaded Ukraine and sent energy prices surging in Europe four years ago, Hetzner did its first price increase in years by 10%.** Hetzner moves first in this market and only when they have to. So, we think these price increases are an indicator of what's about to come in the cloud space. Second, Microsoft is working with AWS to get through GitHub's capacity issues. The two are archenemies in the cloud market. So this is big. Reason: GitHub reported a 10x+ increase in traffic due to agentic workloads. Their COO reported that GitHub served 1 billion commits in 2025. In April, that number was 1.1 billion commits per month. GitHub Actions traffic more than doubled from 2025 to 2026 April. The times are wild. Capacity crunch is real. And it's about to hit the infra space even harder.
-
Enes Çakır liked thisEnes Çakır liked thisUbicloud Postgres is now available on Google Cloud. Claude Code was one of its authors. Three interesting facts: 1/ Ubi in our name comes from ubiquitous. After bare metal and AWS, this is the third platform where we offer managed Postgres. 2/ Ubi PG uses local NVMe. We see 4-5x higher performance than Cloud SQL and AlloyDB. Performance benchmarks to come. 3/ It’s all open source! For this port, we used Claude Code (Opus 4.6) and had OpenAI Codex to do adversarial code reviews. We found that the same AI model tends to agree with itself a lot. When models didn’t agree in code review, we escalated to human to tie-break. What helped us tremendously was comprehensive test coverage. **Claude works best when you can set a verifiable feedback loop - otherwise, you need a human to act as part of the feedback loop and that takes forever.** Ubicloud has 100% line and branch coverage in unit tests. If those pass, Claude starts our E2E test suite, ssh’s into VMs where the tests run, and debugs its code. After Claude completed its project, we still ran the entire code through comprehensive human review. Without this review, we would have completed this project about 10x faster than our AWS port. With it, we were about 20% faster. It’s a trade-off. Interested in trying this out? Link is in the comments. If you find a legitimate bug (one that we fix), we’ll give you 100% off over the next three months. We also won’t blame Claude over it. :)
-
Enes Çakır liked thisEnes Çakır liked thisWe've been building managed Postgres for 15 years. Across all of them, one configuration remained constant: strict memory overcommit. Then, a three-character Linux kernel bug hit us. We had to temporarily disable this setting and figure out why our customers' databases didn't have enough memory. Attached pic shows the fix from Linus. Kudos to Burak Yücesoy for driving this investigation and rolling out the new kernel version in our fleet. In the process, Burak also wrote down all of this together in a nice blog post. - Linux out of memory (OOM) killer - Why PostgreSQL can't tolerate the OOM killer - How Ubicloud solves this issue: fail early, not catastrophically - The three-character Linux kernel bug - How we set the linux commit limits - Our open source implementation One of the most interesting blog posts I've read in a while. Link is in the comments. Please feel free to chime in with any questions and comments. ⬇️⬇️
Experience & Education
-
Ubicloud
View Enes’s full experience
See their title, tenure and more.
or
Licenses & Certifications
-
-
-
-
-
-
From the Big Bang to Dark Energy
Coursera Verified Certificates
-
AstroTech: The Science and Technology behind Astronomical Discovery
Coursera Verified Certificates
-
Volunteer Experience
-
Technology Specialist
Leyla'dan Sonra
- Present 12 years 8 months
Children
Leyla'dan Sonra (After Leila) is an awarded social initiative which makes children with chronic diseases wishes come true.
• Built Children Management System using PHP Laravel framework.
• CMS is used by 2000+ medical students at 45 university hospitals and more than 6000 wishes came true.
• Developed iOS Application for volunteers using Swift. -
Volunteer
Çocuk Vakfı
- Present 26 years 6 months
Children
Projects
-
Biryudumkitap
-
See projectBiryudumkitap is an awarded social initiative that delivers book extracts for free to increase the number of book readers within Turkey.
• Built the mailing web app with AWS integration to send daily email's to +300.000 users using Laravel framework.
• Developed several ad-hoc web apps for internal usage with Laravel and deployed to Digital Ocean.
• Bir Kutu Kitap: Built and maintained the web app for box subscription e-Commerce business using PHP Laravel.
Languages
-
Türkçe
Native or bilingual proficiency
-
İngilizce
Professional working proficiency
Other similar profiles
Explore more posts
-
OBSS
📣New on our Blog! Our Solution Manager Emre Ozan Memiş shares his insights on the future of AI-powered software development with GitHub. Discover how AI is transforming software delivery, empowering developers, and shaping the next era of digital innovation. 🔗Read the full article here: https://lnkd.in/dtwMXxmf #OBSS #GitHub #Microsoft #AI #SoftwareDevelopment #DevOps #FutureOfAI
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content