Prafull Kotecha

Another quality of life pre:invent release - this looks like it will streamline complex state machines. https://lnkd.in/enc2Qsgi

1

A good example of improved security enhancing the functionality of the business instead of impeding the business. The performance testing came back for a backend API call pattern I put together. ☑️ ~85% lower latency ☑️ Lower total cost ☑️ Fewer hops and simpler path ☑️ Reduced attack surface ☑️ Stronger authentication ☑️ Improved auditability & observability ☑️ Only cloud-native components ☑️ Takes workload from more expensive components The only compromise was that some key engineers needed to learn how to set up mTLS between a few specific services, which is also a win IMO.

3

How To Build A Platform Engineering Environment On Kubernetes 👇 ✅ Part 1: https://buff.ly/3u58qhA Part one dives into building the cluster, automating the workflows, and securing the environment. ✅ Part 2: https://buff.ly/47NDG2k Part two dives into platform capabilities like monitoring, observability, and overall cluster management. ✅ Part 3: https://buff.ly/4939GRg Part three dives into various tools and platforms you can use to make platform engineering easier on Kubernetes. #kubernetes #devops #platformengineering

118

1 Comment

Adapting the Simple Sabotage Field Manual (https://lnkd.in/gJq3EzxZ) to modern enterprise IT and Security Architecture: "When possible, escalate all decisions to cross-functional working groups for 'collaboration and alignment.' Ensure these groups include representatives from multiple teams such as security, operations, development, compliance, and legal, and aim for meetings that require full attendance. Additionally, suggest extensive documentation and input from all teams before any decision can be finalized. Use collaborative tools like email threads, project management platforms, and shared documents to track all feedback and changes." Leads to: ❌ Decision-making delays ❌ Overwhelming coordination ❌ Document overload ❌ Loss of accountability ❌ Increased complexity ❌ Silos and mismanagement Instead: ✅ Empower small, cross-functional teams: Grant nimble teams the authority to make the call; Emphasize standardization, not uniformity; Permit deviation from the team's decision only with supporting evidence. ✅ Define clear accountability and ownership over action: Acknowledge that we can't guarantee results; Assign accountability to a single point of contact for gathering input, making informed decisions, and driving execution; Delegate authority to complement accountability. ✅ Streamline communication and tools: Rely on concise, direct communication channels; Use tools that reduce the manual overhead of managing and tracking daily work. ✅ Set strict timelines for feedback and action: Hold managers accountable to communicate their concerns with supporting details early enough to adjust; Expect imperfection and build timelines based on critical path. ✅ Record and communicate decisions: Maintain transparency by documenting decisions and making them available, understandable, and actionable; This is especially critical with standardized architecture patterns and architecture decisions. ✅ Prioritize action over endless collaboration: Require justification for delaying action after a clear majority is reached; Leaders must actively prevent becoming locked in group-based analysis paralysis. ✅ Adopt small, rapid iteration: We are what we repeatedly do, and big goals are accomplished through consistency of habit; small iterations allow for more repetitions and make improvements at each iteration more realistic.

2

Whenever #LangGraph persistence is discussed, it mostly talks about persisting state across nodes and edges throughout the workflow. It can be achieved even by in-memory persistence which is what is mostly talked, documented about. However, there is an additional subtle aspect to it which is to persist it on disk / in DB. It can help an application pick up where it had left off in case of a crash or restart.   I have done a YouTube video on it with details of how it can be done, its nuances, differentiating it from in-memory persistence, and how it can help any AI application. Appreciate any feedback. #GenAI

10

In this post, we take look at some of the key design principles that could be used in Docker, Kubernetes, and other container platforms. https://lnkd.in/eY6nrrq

1

From On-Premises to Multi-Cloud: Yahoo's Kubernetes Transformation Scaling Kubernetes for platforms like Yahoo Sports, Finance, and News is no small feat. Yahoo’s journey from on-premises to multi-cloud architecture is a masterclass in innovation and problem-solving at scale. In a recent talk by Nandhakumar Venkatachalam and Payal Patel, they shared valuable insights into how Yahoo tackled this transformation, balancing complexity, performance, and security. Here are some standout lessons from their experience: 1️⃣ 𝗦𝘁𝗮𝗻𝗱𝗮𝗿𝗱𝗶𝘇𝗶𝗻𝗴 𝘁𝗵𝗲 𝗦𝗲𝘁𝘂𝗽: In-house tools like KDNC and FLEKS simplified managing Kubernetes clusters across environments. These tools leveraged Terraform and CloudFormation to create a consistent, efficient way of provisioning clusters. 2️⃣ 𝗦𝗲𝗰𝘂𝗿𝗶𝗻𝗴 𝗜𝗱𝗲𝗻𝘁𝗶𝘁𝘆 𝗮𝗻𝗱 𝗔𝗰𝗰𝗲𝘀𝘀: Yahoo adopted Athenz and Istio to automate RBAC and establish app/service identities. Security and scalability were seamlessly integrated when using OIDC for EKS or Workload Identity Pools for GKE. 3️⃣ 𝗦𝗰𝗮𝗹𝗶𝗻𝗴 𝗢𝗯𝘀𝗲𝗿𝘃𝗮𝗯𝗶𝗹𝗶𝘁𝘆: With Prometheus, Grafana, and Splunk, Yahoo monitored their infrastructure across 37+ clusters, supporting 135,000 pods, 385,000 containers, and 1.8 million requests per second—a truly massive operation. 4️⃣ 𝗢𝗽𝘁𝗶𝗺𝗶𝘇𝗶𝗻𝗴 𝗡𝗲𝘁𝘄𝗼𝗿𝗸𝗶𝗻𝗴: Yahoo’s approach to cloud-native networking stood out. Custom CIDRs, VPC Peering, and Istio ingress configurations helped reduce costs and improve performance across AWS and GCP regions. 5️⃣ 𝗗𝗿𝗶𝘃𝗶𝗻𝗴 𝗖𝗼𝘀𝘁 𝗘𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝗰𝘆: Bin-packing strategies and resource optimization allowed Yahoo to achieve significant savings. A notable example is their GKE-based Header Bidding project, which reduced costs by 33% while maintaining high throughput. What’s remarkable about Yahoo’s approach is its focus on abstraction—allowing developers to concentrate on building applications while the platform handled the heavy lifting around infrastructure, security, and observability. It’s a model of balancing complexity and usability in a large-scale, multi-cloud environment. You can dive into the full presentation here: https://lnkd.in/eSh77edu

7

Massive news for .NET folks - AWS SnapStart for Lambda is now available for .NET (and Python, but whatever ;) ). I have a personal connection to this release, as I was lucky enough to be a beta tester behind the scenes on SnapStart for .NET. It wasn't just exciting to see the difference it's made to the cold start problem for some of our Lambda stacks (can't wait to get this into production) but it was a fascinating look behind the scenes on how AWS elicits feedback straight from customers just like you and I. I got to work (virtually) side by side with some talented engineering folks from the AWS Lambda/.NET team and understand how they approach experimentation and gathering statistics for the value proposition of whatever feature they're building. There's something exciting about hacking together some C# code into the constructor of your Lambda in a peer programming session with AWS folks from here and abroad. I even got to learn/hear about some of the behind the scenes of how lambda containers are initialised. Valuable insights. Special thanks to our relentlessly awesome solution architect Robbie Cooray for getting me involved in the beta testing, and continuing to do great work behind the scenes. Onwards - to faster .NET Lambdas! https://lnkd.in/g3FzayKw

49

3 Comments

AWS Lambda is an almost instinctual choice when it comes to building on AWS Cloud (and why not? It's awesome!). But are Lambda functions - and the custom libraries and code that comes with them - always the right choice? Particularly for gluing components together? I spend some time exploring the potential of working with native AWS services to produce an event-driven solution with no Lambda or custom code in sight. #aws #eventbridge #software #architecture #serverless

38

6 Comments

In this article I do a deep dive into the most valuable skill for a solution architect: building relationships. If you're an SA, this one skill is your bread and butter.

1

I had the opportunity to read the early release of the book 𝐏𝐨𝐥𝐢𝐜𝐲 𝐚𝐬 𝐂𝐨𝐝𝐞: 𝐈𝐦𝐩𝐫𝐨𝐯𝐢𝐧𝐠 𝐂𝐥𝐨𝐮𝐝-𝐍𝐚𝐭𝐢𝐯𝐞 𝐒𝐞𝐜𝐮𝐫𝐢𝐭𝐲. This book excels in explaining the basics of Policy as Code concepts and delving into how to build effective policies for cloud-native solutions. Given that cloud-native deployments in public clouds are often left insecure, Policy as Code can be a game-changer. If you're looking to deepen your understanding, check out the pre-release through O'Reilly's subscription or pre-order it on Amazon. Highly recommended!

25

📣 [New Blog] One small resource configuration change to use customer-managed KMS key, a giant leap towards secured workflows for customers using AWS Step Functions. As promised, here is the blog and sample application that elaborates the integration of customer-managed AWS KMS key in AWS Step Functions: https://lnkd.in/g3DJqUPj

39

In software engineering, we strive for all the non-functional requirements (the "ilities" of software) to be of very high quality, all the time, no matter the runtime placement or application architecture of software. However, in practice, as seen in normal COTS software....software quality for non-functional requirements is generally proportional to the number of simultaneous users/connections/parallel executions. Edge compute workloads suffer as a result. You see this in practice with IT Crowd's famous quote: "Have you turned it off and on again?"

10

AWS CloudFormation vs. Terraform. Both are great for Infrastructure as Code. CloudFormation? It’s AWS-native, built-in security. Terraform? Cross-platform, multi-cloud flexibility. Which to choose? If you're all-in on AWS, stick to CloudFormation. If you work with different clouds, Terraform wins. Both can automate your infrastructure. Make sure you know when to use each. #InfrastructureAsCode #AWSPractitioner #DevOpsTools

7

6 Comments

On-Prem Kubernetes: 👉🏻 Manage both the Control Planes and Worker Nodes. 👉🏻 Fully customizable. Use any plugin you’d like. Managed Kubernetes: 👉🏻 Manage the Worker Nodes. The Control Planes are managed for you (minus updates). 👉🏻 Less customization. You have to see the world through the vendors eyes. In this episode of the Engineering Revised podcast, Kristina and I break down what you need to know between environments for production. #kubernetes #devops #platformengineering

13

🎉 300 𝗦𝘁𝗮𝗿𝘀 𝗼𝗻 𝗚𝗶𝘁𝗛𝘂𝗯! 𝗧𝗵𝗮𝗻𝗸 𝗬𝗼𝘂! 🌟  I’m thrilled to share that the 𝗠𝘂𝗹𝘁𝗶-𝗔𝗴𝗲𝗻𝘁 𝗢𝗿𝗰𝗵𝗲𝘀𝘁𝗿𝗮𝘁𝗼𝗿 has reached 300 𝘀𝘁𝗮𝗿𝘀 on GitHub! 🚀 This is a significant milestone and a testament to the incredible support and enthusiasm from the developer community.  💡 𝗪𝗵𝗮𝘁 𝗶𝘀 𝗠𝘂𝗹𝘁𝗶-𝗔𝗴𝗲𝗻𝘁 𝗢𝗿𝗰𝗵𝗲𝘀𝘁𝗿𝗮𝘁𝗼𝗿?  It’s a flexible and extensible framework for managing multiple AI agents and handling complex conversations. Whether you’re integrating agents like 𝗔𝗺𝗮𝘇𝗼𝗻 𝗕𝗲𝗱𝗿𝗼𝗰𝗸, 𝗔𝗺𝗮𝘇𝗼𝗻 𝗟𝗲𝘅, 𝗢𝗽𝗲𝗻𝗔𝗜, 𝗟𝗹𝗮𝗺𝗮 𝗔𝗴𝗲𝗻𝘁𝘀, or others, the framework provides adaptability for a wide range of use cases.  📅 𝗥𝗲𝗹𝗲𝗮𝘀𝗲 𝗧𝗶𝗺𝗲𝗹𝗶𝗻𝗲:  🔹 𝗝𝘂𝗹𝘆: Launched the framework in TypeScript.  🔹 𝗔𝘂𝗴𝘂𝘀𝘁: Released the Python version, making the framework accessible to an even broader audience.  ✨ 𝗞𝗲𝘆 𝗙𝗲𝗮𝘁𝘂𝗿𝗲𝘀 𝗜𝗻𝗰𝗹𝘂𝗱𝗲:  ✅ 𝗠𝘂𝗹𝘁𝗶-𝗮𝗴𝗲𝗻𝘁 𝘀𝘂𝗽𝗽𝗼𝗿𝘁: Amazon Bedrock, Amazon Lex, OpenAI, LLama Agents, and more.  ✅ 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝘁 𝗶𝗻𝘁𝗲𝗻𝘁 𝗰𝗹𝗮𝘀𝘀𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻: Dynamically route queries to the best agent.  ✅ 𝗦𝘁𝗿𝗲𝗮𝗺𝗶𝗻𝗴 𝗮𝗻𝗱 𝗻𝗼𝗻-𝘀𝘁𝗿𝗲𝗮𝗺𝗶𝗻𝗴 𝗿𝗲𝘀𝗽𝗼𝗻𝘀𝗲𝘀: Flexibility for real-time and batch workflows.  ✅ 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗺𝗮𝗻𝗮𝗴𝗲𝗺𝗲𝗻𝘁: Maintain coherent multi-turn interactions.  ✅ 𝗘𝘅𝘁𝗲𝗻𝘀𝗶𝗯𝗹𝗲 𝗮𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲: Easily integrate custom agents.  🚀 𝗘𝘅𝗮𝗺𝗽𝗹𝗲𝘀 & 𝗤𝘂𝗶𝗰𝗸 𝗦𝘁𝗮𝗿𝘁: Get hands-on experience with the Multi-Agent Orchestrator through our diverse set of examples:  🔹 𝗥𝗲𝗮𝗱𝘆-𝘁𝗼-𝗿𝘂𝗻 𝗦𝗰𝗿𝗶𝗽𝘁𝘀: Start locally with standalone scripts in both Python and TypeScript.  🔹 𝗗𝗲𝗺𝗼 𝗔𝗽𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻𝘀:  - Chat Demo App: Specialized agents for travel, weather, math, and health.   - E-commerce Support Simulator: AI-powered customer support with automation and human-in-the-loop capabilities.  🔹 𝗦𝗮𝗺𝗽𝗹𝗲 𝗣𝗿𝗼𝗷𝗲𝗰𝘁𝘀: Explore a variety of projects, including chat interfaces, AI-powered customer support systems, FastAPI integrations, and tools for converting natural language into structured data—all available in the examples folder.  Anthony Bernabeu and I are the co-authors and active maintainers of this project. We’re deeply committed to its growth and ensuring it continues to deliver value to the community.  🙏 𝗧𝗵𝗮𝗻𝗸 𝗬𝗼𝘂! To everyone who has contributed, supported, or provided feedback—you’ve made this journey incredibly rewarding.  ✨ 𝗜 𝗰𝗮𝗻’𝘁 𝘄𝗮𝗶𝘁 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗻𝗲𝘅𝘁 𝗺𝗶𝗹𝗲𝘀𝘁𝗼𝗻𝗲: 1000 𝘀𝘁𝗮𝗿𝘀! 🌟  Let’s make it happen together! If you haven’t already, check it out, give it a ⭐️, and let me know how you’re using the framework or what use cases you’d love to see.  https://lnkd.in/gcfSYJHM #opensource #AI #GitHub #MachineLearning #MultiAgentSystems #AmazonBedrock #OpenAI #multiagentorchestrator

47

16 Comments