According To Ali

GPUs are a critical yet expensive commodity for AI workloads. Justifying their high cost requires maximizing utilization. To keep things in focus, we are not discussing token consumption or tokeneconomics here; This is about GPU optimisation (at the infra and platform). In this post I will describe an approach to building an optimized GPU-as-a-Service using the below key components: GPU Sharing : Increases workload density and optimising GPU usage. Model As A Service (MaaS) : Standardizes and centralizes model deployment and consumption. Kueue : Controls prioritisation, duration and placements of GPU workloads. Self serve : Enables different roles to self serve to utilise GPUs available within the organisation for usecases. GPU Sharing + MaaS + Kueue + Self-Serve = GPU-as-a-Service Solution overview: The big picture: The above goal translates to below high level overview: Demonstration of GPU-as-a-Service in action: Let's dive into the bits now. GPU Sharing: GPUs were des...

Managing edge devices has been a complex process as traditional IT ops tools fall short in distributed, low-connectivity environment to manage huge quantity of devices.  Red Hat Edge Manager  (Open source project: FlightControl , GA'd by Red Hat on late Jan, 2026) solves these challenges by providing streamlined management of edge devices and applications through a declarative approach . Now, there's a fair bit to unpack here. But for simplicity this is how I am going to map those 3 things here: Management of edge devices: I am mapping this to LCM (including upgrade, patch etc) of the underlying OS (in this case RHEL OS of BootC flavor or at least UBI based RHEL ). Managing applications: Mapping this to deploying applications and LCM of the applications stack on the OS. Declarative approach: This one is super interesting. To me this is very K8s-yy but in the world of edge devices running linux (RHEL OS, as of today). And then this thing also has MCP : This is my next prob...

I want to fetch my secrets from Azure KV and I don't want to use any password for it. Let's see how this can be implemented. This is yet another blog post (YABP) about ESO and Azure Workload Identity. Why Passwordless Auth: It is a common practice to use some sort of "master password" (spn clienid, clientsecret etc) to access Secret Vaults (in this case it is AZ KV) but that master password becomes a headache to manage (rotate, prevent leak etc). So, the passwordless auth to AKV is ideal.  Why ESO: This is discussed and addressed in the conclusion section. Workload Identity (Passwordless Auth): Lets make a backward start (just for a change). I will try to explain how the passwordless auth will work. This will make more sense when you will read through the detailed implementation section. Here's a sequence diagram to explain it: There's no magic here. This is a well documented process by microsoft  here . The below diagram (directly copied from the official doc...

It is mid-2025 and the cogs of AI are at full speed. So we (I and Mobin) decided to do our own AI project. We called it "IntelliLogs".  IntelliLogs at a glance: Demo:  https://www.youtube.com/watch?v=OXMlORwyMQk In this post I will describe why we did what we did, what is it that we did and how we did it. I will share my personal experience. I am hoping this will, at least, be an interesting read. Table of contents: Why IntelliLogs What is IntelliLogs How IntelliLogs was developed Future of IntelliLogs Conclusion References Why IntelliLogs: Personal motivation 💪 to this were: Explore and experience what does an AI app look like from an architectural and engineering perspective Explore the realm of Huge LLMs (eg: GPT-4.1-170B,  Gemini Pro etc) vs small LLMs (eg: granite-7b, gemma-4b) Explore the possibilities of model tuning / making a model without being a data scientist. How easy or hard it is, what tools available etc. We also wanted to tackle a "not too far from ...