Puja Abbassi
@puja.dev
VP Product @GiantSwarm; CNCF Ambassador; CKA; recovering Data Science Academic; into Chinese Martial Arts; not updating his website puja.dev
Reposted by Puja Abbassi
AI infra just got a standard. CNCF launched the Kubernetes AI Conformance Program, and @puja.dev was part of the group that helped define it.
Giant Swarm is now one of the first certified platforms.
Giant Swarm is now one of the first certified platforms.
Infrastructure for AI is finally getting a standard » Giant Swarm
AI infrastructure just got simpler. Giant Swarm is officially CNCF-certified to run AI/ML workloads on Kubernetes.
www.giantswarm.io
November 12, 2025 at 9:01 AM
AI infra just got a standard. CNCF launched the Kubernetes AI Conformance Program, and @puja.dev was part of the group that helped define it.
Giant Swarm is now one of the first certified platforms.
Giant Swarm is now one of the first certified platforms.
And best thing, it’s open source - Streamline Complex #AI Inference on #Kubernetes with #NVIDIA Grove | NVIDIA Technical Blog
Streamline Complex AI Inference on Kubernetes with NVIDIA Grove | NVIDIA Technical Blog
Over the past few years, AI inference has evolved from single-model, single-pod deployments into complex, multicomponent systems. A model deployment may now consist of several distinct components…
developer.nvidia.com
November 11, 2025 at 4:04 PM
And best thing, it’s open source - Streamline Complex #AI Inference on #Kubernetes with #NVIDIA Grove | NVIDIA Technical Blog
A bit older but still important to define #contextengineering and introduce some techniques for optimization - Effective context engineering for #AI agents www.anthropic.com/engineering/...
Effective context engineering for AI agents
Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.
www.anthropic.com
November 11, 2025 at 2:13 PM
A bit older but still important to define #contextengineering and introduce some techniques for optimization - Effective context engineering for #AI agents www.anthropic.com/engineering/...
It’s really cool seeing this outperform models with tokenizers, especially in non-English use cases, recommend reading the research background - Introducing TFree-HAT 7B: Tokenizer-Free Models Achieving Top-Tier Multilingual Performance aleph-alpha.com/introducing-... #AI #LLM #Research
Introducing TFree-HAT 7B: Tokenizer-Free Models Achieving Top-Tier Multilingual Performance - Aleph Alpha
We’re excited to share something fresh from our research lab. Just recently, we made public two new tokenizer-free models built on our Hierarchical Autoregressive Transformer (HAT) architecture:…
aleph-alpha.com
October 9, 2025 at 10:45 AM
It’s really cool seeing this outperform models with tokenizers, especially in non-English use cases, recommend reading the research background - Introducing TFree-HAT 7B: Tokenizer-Free Models Achieving Top-Tier Multilingual Performance aleph-alpha.com/introducing-... #AI #LLM #Research
Reposted by Puja Abbassi
"The State of OCI Artifacts for AI/ML" www.gorkem-ercan.com/p/the-state-...
The State of OCI Artifacts for AI/ML
I’ve spent the last 18 months watching OCI artifacts for AI/ML go from “interesting idea” to production-grade infrastructure in enterprises.
www.gorkem-ercan.com
October 7, 2025 at 2:30 PM
"The State of OCI Artifacts for AI/ML" www.gorkem-ercan.com/p/the-state-...
Nice to see some alternative approaches to improving coding LLMs - CWM: An Open-Weights LLM for Research on Code Generation with World Models ai.meta.com/research/pub... #AI #research
CWM: An Open-Weights LLM for Research on Code Generation with World Models | Research - AI at Meta
We release Code World Model (CWM), a 32-billion-parameter open-weights LLM, to advance research on code generation with world models. To improve code...
ai.meta.com
October 7, 2025 at 3:23 PM
Nice to see some alternative approaches to improving coding LLMs - CWM: An Open-Weights LLM for Research on Code Generation with World Models ai.meta.com/research/pub... #AI #research
Good overview of memory solutions for #LLM - #AgenticAI: Implementing Long-Term Memory towardsdatascience.com/agentic-ai-i...
Agentic AI: Implementing Long-Term Memory | Towards Data Science
The problem and current solutions
towardsdatascience.com
September 12, 2025 at 10:45 AM
Good overview of memory solutions for #LLM - #AgenticAI: Implementing Long-Term Memory towardsdatascience.com/agentic-ai-i...
I like how they are attacking a common problem with legacy software with a specialized agent - GitHub Copilot app modernization for Java github.blog/changelog/20... #AI
GitHub Copilot app modernization for Java now in public preview - GitHub Changelog
We’ve released the public preview of GitHub Copilot app modernization for Java—an AI-powered solution designed to simplify and accelerate Java upgrades and cloud migrations. By combining the…
github.blog
September 9, 2025 at 3:04 PM
I like how they are attacking a common problem with legacy software with a specialized agent - GitHub Copilot app modernization for Java github.blog/changelog/20... #AI
This is not yet another context engineering guide, this is actually technical and gives best-practices around what to be careful around context when building agents both to get good results but also to be cost effective manus.im/blog/Context...
Context Engineering for AI Agents: Lessons from Building Manus
This post shares the local optima Manus arrived at through our own "SGD". If you're building your own AI agent, we hope these principles help you converge faster.
manus.im
September 9, 2025 at 10:45 AM
This is not yet another context engineering guide, this is actually technical and gives best-practices around what to be careful around context when building agents both to get good results but also to be cost effective manus.im/blog/Context...
Indeed, they are not, and they don't need to be to be useful. Still, good to remind yourself that they are not and that they have significant limitations. - LLMs aren’t world models yosefk.com/blog/llms-ar...
LLMs aren’t world models
August 10th, 2025
yosefk.com
September 4, 2025 at 3:04 PM
Indeed, they are not, and they don't need to be to be useful. Still, good to remind yourself that they are not and that they have significant limitations. - LLMs aren’t world models yosefk.com/blog/llms-ar...
I kinda liked the idea of staying with MCP but behind the MCP not having the tool directly, but a specialized agent that works with the tool and offers the benefits of an agent (which the blog post also lists) devblogs.microsoft.com/blog/can-you...
Can You Build Agent2Agent Communication on MCP? Yes! - Microsoft for Developers
MCP has evolved significantly beyond its original goal of “providing context to LLMs.” With recent enhancements including resumable streams, elicitation, sampling, and notifications (progress and…
devblogs.microsoft.com
September 4, 2025 at 10:45 AM
I kinda liked the idea of staying with MCP but behind the MCP not having the tool directly, but a specialized agent that works with the tool and offers the benefits of an agent (which the blog post also lists) devblogs.microsoft.com/blog/can-you...
I wouldn't do all of those, but generally the question of "how do I practice my craft" is a valuable one marginalrevolution.com/marginalrevo...
How I practice at what I do - Marginal REVOLUTION
Following up on my post a few days ago, about the value of deliberate practice for knowledge workers, a number of you asked me what form my practice takes. A few of you were skeptical, but it is…
marginalrevolution.com
September 3, 2025 at 3:05 PM
I wouldn't do all of those, but generally the question of "how do I practice my craft" is a valuable one marginalrevolution.com/marginalrevo...
Good short post about speaking out - Are you stuck in movie logic?
Are you stuck in movie logic?
Consider just saying what the problem is
usefulfictions.substack.com
September 1, 2025 at 10:45 AM
Good short post about speaking out - Are you stuck in movie logic?
Reposted by Puja Abbassi
It would be great to have a Kubernetes icon in Font Awesome. Upvote this request to bump it up the priority list | Icon request: Kubernetes-logo · Issue #13331 · FortAwesome/Font-Awesome #SuggestedRead #devopsish github.com/FortAweso...
Icon request: Kubernetes-logo · Issue #13331 · FortAwesome/Font-Awesome
What can it be used for? Example image (optional) Icon request checklist [X ] This is a single icon or matched pair (Ex: lock / unlock) [ X] The title starts with "Icon request: " and is ...
github.com
July 24, 2025 at 12:50 AM
It would be great to have a Kubernetes icon in Font Awesome. Upvote this request to bump it up the priority list | Icon request: Kubernetes-logo · Issue #13331 · FortAwesome/Font-Awesome #SuggestedRead #devopsish github.com/FortAweso...
Reposted by Puja Abbassi
Dominik is very much looking forward to giving this talk!
July 21, 2025 at 2:18 PM
Dominik is very much looking forward to giving this talk!
Reposted by Puja Abbassi
Reposted by Puja Abbassi
AI agents have brand “preferences” and are attracted to different kinds of ads (right now, Operator is a fan of buying whatever Bing advertises, Claude has different preferences)
There is likely going to be a lot of money spent trying to influence this in the nearish future.
There is likely going to be a lot of money spent trying to influence this in the nearish future.
June 29, 2025 at 11:08 PM
AI agents have brand “preferences” and are attracted to different kinds of ads (right now, Operator is a fan of buying whatever Bing advertises, Claude has different preferences)
There is likely going to be a lot of money spent trying to influence this in the nearish future.
There is likely going to be a lot of money spent trying to influence this in the nearish future.
Reposted by Puja Abbassi
Not satisfied with ONE talk, our own @puja.dev is teaming up with the fantastic @wiggitywhitney.bsky.social for another talk this time at OpenSSF Community Day.
Ozzie & Nova - Supply Chain Shenanigans: A Kubernetes Security Play About OpenSSF Standards
openssfcdjapan2025.sched.com/event/23VH0/
Ozzie & Nova - Supply Chain Shenanigans: A Kubernetes Security Play About OpenSSF Standards
openssfcdjapan2025.sched.com/event/23VH0/
OpenSSF Community Day Japan 2025: Ozzie & Nova - Supply Chain Shenanigans:...
View more about this event at OpenSSF Community Day Japan 2025
openssfcdjapan2025.sched.com
June 17, 2025 at 6:43 AM
Not satisfied with ONE talk, our own @puja.dev is teaming up with the fantastic @wiggitywhitney.bsky.social for another talk this time at OpenSSF Community Day.
Ozzie & Nova - Supply Chain Shenanigans: A Kubernetes Security Play About OpenSSF Standards
openssfcdjapan2025.sched.com/event/23VH0/
Ozzie & Nova - Supply Chain Shenanigans: A Kubernetes Security Play About OpenSSF Standards
openssfcdjapan2025.sched.com/event/23VH0/
Hope to meet some of you there and if you're not coming to Tokyo keep your eyes out for the recording on YouTube very soon!
Attending #KubeConCloudNativeCon in Japan? 🇯🇵
Well, our very own @puja.dev is!
On Monday June 16th, he’ll be talking about:
“Platform Engineering Day 2: Why Service Iterations are the Crux of Developer Platforms”
Stop by and say hey 👋
kccncjpn2025.sched.com/event/1x70r/...
Well, our very own @puja.dev is!
On Monday June 16th, he’ll be talking about:
“Platform Engineering Day 2: Why Service Iterations are the Crux of Developer Platforms”
Stop by and say hey 👋
kccncjpn2025.sched.com/event/1x70r/...
KubeCon + CloudNativeCon Japan 2025: Platform Engineering Day 2: Why Service...
View more about this event at KubeCon + CloudNativeCon Japan 2025
https://kccncjpn2025.sched.com/event/1x70r/platform-engineering-day-2-why-service-iterat[…]-the-crux-of-developer-platforms-puja-abbassi-giant-swarm
June 12, 2025 at 10:14 AM
Hope to meet some of you there and if you're not coming to Tokyo keep your eyes out for the recording on YouTube very soon!
Reposted by Puja Abbassi
We’re excited to introduce Text-to-LoRA: a Hypernetwork that generates task-specific LLM adapters (LoRAs) based on a text description of the task. Catch our presentation at #ICML2025!
Paper: arxiv.org/abs/2506.06105
Code: github.com/SakanaAI/Tex...
Paper: arxiv.org/abs/2506.06105
Code: github.com/SakanaAI/Tex...
June 12, 2025 at 1:47 AM
We’re excited to introduce Text-to-LoRA: a Hypernetwork that generates task-specific LLM adapters (LoRAs) based on a text description of the task. Catch our presentation at #ICML2025!
Paper: arxiv.org/abs/2506.06105
Code: github.com/SakanaAI/Tex...
Paper: arxiv.org/abs/2506.06105
Code: github.com/SakanaAI/Tex...
Shout out to the one and only @castrojo.bsky.social. I just installed my first projectbluefin.io on a miniPC I had lying around and I gotta say it's the first time in a while I've been enjoying a Linux distro. Everything makes tons of sense and the defaults are great 🙏
Bluesky
castrojo.bsky.social
May 24, 2025 at 6:24 PM
Shout out to the one and only @castrojo.bsky.social. I just installed my first projectbluefin.io on a miniPC I had lying around and I gotta say it's the first time in a while I've been enjoying a Linux distro. Everything makes tons of sense and the defaults are great 🙏
Reposted by Puja Abbassi
Reposted by Puja Abbassi
Shh, hey
Want a secret mobile app feature that gives you the option to change the app icon into a messy dark-mode-with-white-border variant?
Tap on the card below ⬇️
(Note: on web, this just shows an error page. This needs to be in-app; tested on Android)
Want a secret mobile app feature that gives you the option to change the app icon into a messy dark-mode-with-white-border variant?
Tap on the card below ⬇️
(Note: on web, this just shows an error page. This needs to be in-app; tested on Android)
bsky.app
May 14, 2025 at 6:19 AM
Shh, hey
Want a secret mobile app feature that gives you the option to change the app icon into a messy dark-mode-with-white-border variant?
Tap on the card below ⬇️
(Note: on web, this just shows an error page. This needs to be in-app; tested on Android)
Want a secret mobile app feature that gives you the option to change the app icon into a messy dark-mode-with-white-border variant?
Tap on the card below ⬇️
(Note: on web, this just shows an error page. This needs to be in-app; tested on Android)