Hari Prabhakaran

Introducing Strands Agents 1.0: Build production-ready, multi-agent systems in a few lines of code 🤖🔄🚀https://go.aws/44UPj8i An #opensource SDK, Strands Agents takes a model-driven approach to building & running AI agents in just a few lines of code. Strands scales from simple to complex agent use cases, & from local development to deployment in production. Read the blog to learn more. #AWS #generativeAI

826

15 Comments

In-memory databases are the architectural backbone of real-time AI. They’re a necessary advantage for memory-intensive workloads that rely on instant access to context. Modern systems like Redis blend durability, multi-model flexibility (JSON, vectors, time-series), and tiered storage to deliver the speed of memory with the reliability of disk. The result? Real-time analytics, instant personalization, and AI systems that actually feel intelligent. Here’s how we deliver the speed of memory with the reliability of disk: https://lnkd.in/gSm65kdB

49

“1 million rows per second.” In his opening keynote, our CEO Max Liu walked through what that really means in production: - TiDB processing 1M rows/sec - 10x higher throughput for cross-region replication - A data layer built to make multi-tenant architectures actually practical at scale If you’re thinking about SaaS, multi-region, or AI-heavy workloads, this is the kind of infrastructure ceiling you want. 🎥 Watch the full keynote from Max Liu at #TiDBSCaiLE to see what’s behind the numbers: https://ow.ly/BgKl50YetBq #TiDB #DistributedSQL #MultiTenancy #CloudNative #Scalability

24

Over the past 4 months, under the leadership of Lisa Su, Anush E., & Vamsi Boppana, AMD has made rapid progress & increased the sense of urgency by 10x 🚀 They now have CI/CD, access to internal development nodes, & plans to make an community developer cloud. Still lots of progress ahead, SemiAnalysis engineering will continue engage with AMD to increase their probability of success. Link to full writeup in comments 👇

494

7 Comments

🤖🧠 Redis x LangGraph Memory The new langgraph-checkpoint-redis integration enables AI agents with persistent memory, featuring thread-level preservation and cross-thread memory with vector search capabilities. Enhance your AI agents with advanced memory - check it out! 👉 https://lnkd.in/gkE5VVSy

654

25 Comments

We go inside Amazon's AI chip lab in Austin, Texas, where the AWS Annapurna Labs unit designs, improves and engineers the chip to the server. Bloomberg's Edward Ludlow gets hands-on with Trainium2 ahead of a big data center launch bloom.bg/3GxYnHH

628

5 Comments

Advancing DevOps with #Amazon Nova. 👨‍💻💻💡 https://go.aws/46UNwT6 Amazon Nova are foundation models that deliver frontier intelligence and industry-leading price-performance. Eric Avery, Sr. Director of IT and Architecture at Sumo Logic, shares how the team integrated Amazon Nova Lite into its Mobot AI-powered assistant to accelerate investigations and troubleshooting. With Nova Lite, Sumo Logic achieved a 24% reduction in costs and a 95% improvement in result accuracy #AWS

173

19 Comments

Building AI infrastructure on-prem? Here are foundational principles to keep in mind: Modular architecture: Decouple compute from storage so each can scale independently. This makes it easier to adapt to growing or shifting workloads over time. Future-ready hardware: Select GPUs and CPUs not just for current workloads but with an eye on scalability, including support for newer accelerator types. Scalable design: Ensure the system allows seamless addition of compute nodes or storage without a full redesign. Open ecosystems: Rely on open source tools and open standards to stay agile, reduce long-term costs, and avoid vendor lock-in. This helps you tailor the stack as your needs evolve. With the right foundation, on-prem setups offer deep control, performance tuning, and long-term flexibility. More tips? Full video on YouTube: https://lnkd.in/eXV6_A6r #Canonical #Ubuntu #OpenSource #Cloud #AI

34

2 Comments

Anyscale 🤝GKE Through our continued collaboration with the Google Kubernetes Engine (GKE) team, we’re advancing the experience to scale Ray and Kubernetes in the cloud. This work includes: Label Selectors for improved scheduling: https://lnkd.in/gUZ8iVUW New Features for AI Scaling: https://lnkd.in/gAGSMiir Enhanced TPU Support: https://lnkd.in/gsKDvwpv

65

3 Comments

Nvidia CEO Jensen Huang on Wednesday dismissed concerns of diminished returns in AI because it will "forever be about thinking," making infrastructure buildout essential. Read more: cnb.cx/4nZNAHr

211

22 Comments

At Amazon's AGI Lab, XJ Wang's Reinforcement Learning Pod team trains AI agents using RL gyms for real-world applications. His team handles everything from model training to product integration across Amazon. Learn more:

65

2 Comments

165,000 resources across 625 workspaces and 38 AWS accounts, supporting 170 engineers. That's automation with HCP Terraform. 🪄 Watch to learn more about Benchling's journey to a unified infrastructure as code (IaC) approach with HashiCorp: https://lnkd.in/gegr69x5

49

Longhorn is a distributed block storage system for Kubernetes. Longhorn creates a dedicated storage controller for each block device volume and synchronously replicates the volume across multiple replicas stored on various nodes. More: https://ku.bz/vb5C-vQQ6

202

6 Comments

We ran a head-to-head benchmark of AWS Textract against nanonets-ocr-2, our VLM-based doc processing model. Textract breaks reading order on multi-column pages. It skips charts and maps entirely. It struggles with merged cells and nested tables. And it still outputs raw text blocks instead of structured markdown. If your RAG pipeline or agent is producing garbage answers, the problem probably isn't your LLM. It's your parser. Here's what we found across 7 dimensions: - Layout and reading order: Textract gives up on complex pages. We preserve every section in order. - Visual elements: Textract labels them "Figure 1" and moves on. We describe the content. - Complex tables: Merged cells, spanning rows, sub-headers. We handle them. Textract doesn't. - Multilingual: 200+ languages out of the box. No config. - Signatures and stamps: Detected, linked to signees, included in output. - Accuracy on hard docs: +15% better on Docext and OlmOCR-Bench. - LLM-ready output: Clean markdown with doc structure preserved, and confidence scores One model. Structured markdown out. Confidence scores included. Deploys on-prem in a single Docker container. Try it on the documents your current parser can't handle.

21

🚀 The Responsiveness Mandate: Scaling GenAI with Redis & AWS What does it really take to move Generative AI from pilot to production, and keep it fast, scalable, and reliable in the real world? At an exclusive, invite-only roundtable hosted at Conrad Bengaluru, Redis and AWS brought together 11 senior technology leaders from some of India’s most innovative companies to explore that question. The conversation went deep tackling: ⚡ The non-negotiables of real-time responsiveness in GenAI 📈 Scaling systems without breaking under peak load ⚙️ Balancing model complexity with performance, cost, and latency 🛠️ Infrastructure decisions shaping next-gen intelligent applications The evening opened with a guided coffee cupping and collaborative blend creation, using coffee attributes like freshness, latency, complexity, and finish as a creative lens to unpack the challenges of real-time AI systems. The discussions spanned real-world deployment challenges, from handling unpredictable traffic spikes to optimising retrieval for large context windows, revealing how different industries are approaching responsiveness in their own GenAI journeys. #GenAI #EnterpriseTech #YourStoryEvents #AWS #Redis #TheResponsivenessMandate #Roundtable #AIInfrastructure #TechLeadership Sumeet Singhal | Prabhash Mishra | Debayan Bose | Charu Venkatraman | Karthik Jaya | Ankit Srivastava | Gaurav Konar | Anshul Sharma | Karthikeyan Ramasamy | Chaitanya Bharadwaj | Narayan Babu | Nihal Singh | Jitender Singh Gahlot | Avinash Venkatagiri [In partnership with Redis and Amazon Web Services (AWS)]

19

1 Comment

As AI grows more specialized, infrastructure must keep up. F5 and NVIDIA are helping enterprises build scalable, secure AI factories with BIG-IP Next for Kubernetes on BlueField-3 DPUs—enabling faster inference, smarter prompt routing, and better performance for LLMs. Explore how we're supporting the shift to purpose-built AI: https://go.f5.net/q3cy5o44 #F5 #GTCParis #NVIDIA #Kubernetes #LLM #AgenticAI

199

3 Comments

AI workloads are growing fast and so are the operational challenges. At Fully Connected 2025, Robert Nishihara, Co-founder of Anyscale, shared a clear path forward: a unified stack of Kubernetes, Ray, PyTorch, and vLLM. His talk walks through live demos featuring autoscaling GPU pools and sub-10ms generation latency, showing what’s possible when these tools work together. The full session is now available on demand. Worth a watch: https://lnkd.in/gUZHC8hJ 

54

1 Comment

Kikoff is constantly innovating to build smarter, faster financial tools. In an interview with Built In SF, Engineering Manager Patrick Wang shares how the team is tackling technical debt and optimizing performance to solve problems for customers, faster. Read more on what’s ahead: https://lnkd.in/gBvk7cQ9

10

When video engineers talk about latency, jitter, and workload predictability, they’re often really talking about the invisible tax of legacy architecture. In this episode of Voices of Video, Sean Varley (Chief Evangelist at Ampere) breaks down why traditional x86 platforms - built for general-purpose computing - are fundamentally misaligned with modern video workflows. He gets into how multithreading, power scaling, and turbo boosting often create more problems than they solve. Why? Because video encoding needs predictability, not variability. The magic of Ampere’s ARM-based processors isn’t necessarily about energy efficiency (though that’s impressive). It’s about single-threaded cores that behave like adults in the room - no noisy neighbors, no jittery processing, no half-baked turbo tricks. You get consistent, deterministic execution. That’s critical when real-time performance matters. If you’ve ever had to disable hyperthreading just to get your system to behave during a livestream, or if your current stack feels like a Frankenstein of tuning hacks, you’ll want to hear Sean’s take. 🎧 Listen to the full episode: https://netint.biz/4laGG06 #VideoStreaming #ARMArchitecture #CloudComputing #VoicesOfVideo #EncodingTech #MediaEngineering | NETINT Technologies | Ampere |

35

1 Comment