Activity
7K followers
-
Simba Khadder reposted thisSimba Khadder reposted thisIf you've tried setting up Hermes or OpenClaw, you quickly learn that designing memory/context is extraordinarily complicated. Cassidy Williams' blog mentions Redis as the "boring" infra answer to the context puzzle, reminding me of the terrific talk Simba Khadder gave at AI Council this year. His talk introduces "Context Engineering 2.0" - a systems-level approach to building agentic AI where context is treated as first-class infrastructure. Scalable agentic systems require a context engine. Useful resource: https://lnkd.in/eaxeNtZ3
-
Simba Khadder reposted thisSimba Khadder reposted thisWhen most people hear "agent memory," what comes to mind is personalization.. but there's another piece that gets less attention. Let's take a walk with Simba Khadder, who leads Context Engine at Redis, as he explains. It comes down to the agent being stateful, making decisions based on what it's already done, not just what it knows about you. ➡️ A coding agent that remembers which file or module to touch to ship a new feature ➡️ A support agent that recognizes a problem it's seen before and remembers the ticket it already filed Personalization is powerful on its own. But this is the difference between an agent that personalizes and one that also compounds: it learns from its own mistakes and gets better run over run. The way this typically works under the hood: as the agent works, important information gets extracted from the conversation and stored outside the model as long-term memory. When it's needed again, it's retrieved by semantic similarity plus metadata filters, while the session itself gets summarized and trimmed instead of replaying the full history every time. Redis has something called Redis Iris, their real-time context engine for agents. Agent Memory is one of the pieces inside it: it manages both short-term session state and long-term memory that carries across tasks, so an agent has a persistent record to work from instead of starting from zero every run. Memory that compounds is what turns an agent from a tool you use into one that gets better because you used it.
-
Simba Khadder shared thisWe've been working on something really big at Redis, and it's almost time to share. I'll be speaking alongside our CEO Rowan Trollope on Monday, May 18 at 10am PT about how we're solving the largest problem in the agent space: Context. Context is truly all you need. You're not going to want to miss this. https://lnkd.in/gX5irT3rSimba Khadder shared thisMillions of engineers already build with Redis. Now you can bring your agents to production on Redis. Join Redis CEO, Rowan Trollope, and Director of Engineering & Head of AI product, Simba Khadder, for a special livestream Monday, May 18 at 10am PT.
-
Simba Khadder reposted thisSimba Khadder reposted thisJust 1️⃣ day to OpenXData live virtual conference. Hot takes on Data, Lakehouse, AI pipelines, Query engines, LLMs, and Context Engineering, right from engineers building these. Some standout sessions I’m looking forward to👇 🔍 Simba Khadder talks about building a context engine — making data pipelines agent-ready. 🤖 Tosh Rayadhurgam dives into non-deterministic query layers and what agents break in today’s architectures. ⚙️ Suman Debnath breaks down the physics of LLM inference — where latency actually comes from. 🚀 Kyle Weller+ Rui Mao 's talks explore the post-JVM world — native, GPU-accelerated query engines reshaping the lakehouse. 🧠 Maxime Beauchemin walks through building an AI data agent that replaces large parts of a traditional data team. 🤖 Rahil Chertara and Timothy Brown talk about unstructured/AI data storage in Apache Hudi ⛸️ Julien Le Dem discusses columnar storage evolution in the AI era and what it means for Apache Parquet. 🧱 Will Manning presents GPU-native columnar storage (Vortex) — rethinking formats for AI + analytics. 🧬 Chang She covers managing multimodal data at exabyte scale for AI training pipelines (LanceDB). 📦 Ruiyang Wang shows why PDF pipelines are an attack surface — and how to build them safely. 📊 Yufei Gu and Kevin Liu explains how open catalogs + datasets unlock real interoperability across engines, for Hudi and Apache Iceberg. 🧵 A wide range of talks from the Apache Hudi ecosystem (JD, Uber/Xinli shang, Conductor) — from real-world scale pipelines to evolving Hudi for vector search, unstructured data, and AI-native lakehouses. If you care about where data systems are actually heading, this lineup is 🔥 See you all there. Grab your free seat here 👉 https://www.openxdata.ai #OpenXData #DataEngineering #Lakehouse #AI #AIAgents #ContextEngineering #DataInfrastructure #OpenSource
-
Simba Khadder shared thisFrom Data Lake to Context Lake, I'll be speaking at the Onehouse OpenXData conference on Apr 29th at 10a PT about Context Engineering from a Data Engineering lens. It's a virtual conference, so check it out. Hope to see you there!
-
Simba Khadder shared thisRedis is hosting an AI office hours. Come hang out and get your questions answered by both Andrew Brookins and myself. Hope to see you there! https://lnkd.in/gU-B-qjd
-
Simba Khadder shared thisSpeaking at AI Council all about context engineering. If you've been working agents, you already know, Context is all you need! I've been going for years and have always loved it. This is my first time going as part of Redis and I'm stoked to share more about what we're up to.
-
Simba Khadder reposted thisSimba Khadder reposted thisSimba Khadder leads AI strategy at Redis. He joins Kevin Ball to discuss context engines, agentic AI, memory systems, and how engineering teams adapt as AI development accelerates. https://lnkd.in/g9P_UKz9Unlocking the Data Layer for Agentic AI with Simba Khadder - Software Engineering DailyUnlocking the Data Layer for Agentic AI with Simba Khadder - Software Engineering Daily
-
Simba Khadder shared thisMost people think of acquisitions as the end state. With Featureform, it was just the start of a new chapter. Today, we're releasing Redis Feature Form, a new enterprise-grade feature store platform that's fundamentally different than what we started with. Over the past few months, the team re-architected Featureform and evolved it into a core part of Redis. We took a proven product and transformed it to work at enterprise scale: multi-tenant, enhanced RBAC, deeply integrated with the data layer, with real-time and streaming capabilities built in. Feature stores sit right in the critical path of production ML, where iteration speed, latency, and reliability make or break you. Redis Feature Form is one platform for defining, managing, and serving features. Built for real, enterprise workloads. We're iterating faster than ever, and the demand has been incredible. If you're running ML in production or stuck maintaining an aging internal feature store, let's chat.Simba Khadder shared thisToday, we introduced Redis Feature Form, a new enterprise-grade feature store platform for production machine learning. Feature Form is a big step forward in how we show up for production ML. For years, we’ve been part of the serving layer for real-time features. Now we’re moving higher in the stack, giving ML teams a managed system to: ➡️ Define features ➡️ Orchestrate pipelines ➡️ Track lineage ➡️ Serve features with sub-millisecond latency. Production ML pain isn’t about the model. It’s about getting features into production reliably, keeping training and inference aligned, and avoiding a mess of custom pipelines that are expensive to maintain. Feature Form is built for teams already running real workloads in fraud, risk, recommendations, and personalization, so feature infrastructure is easier to run, easier to govern, and faster in production. Read more about Feature Form from TechTarget here: https://lnkd.in/gBWztgSMRedis unveils Feature Form to improve AI, ML workloads | TechTargetRedis unveils Feature Form to improve AI, ML workloads | TechTarget
Simba Khadder replied to a comment
2mo
Demetrios Brinkmann end of an era and the start of a new one. Let’s surf soon
-
Simba Khadder liked thisSimba Khadder liked thisReally enjoyed the “Rethinking the Semantic Layer” conversation with Jacopo Tagliabue, Paul Blankley, Matthew Mullins and Simba Khadder. It gets at a tension I’m hearing more and more as from Snowflake customers building agentic data systems: how do we give agents the flexibility to answer complex, long-tail questions without giving up trust and consistency? Agents are getting remarkably good at writing SQL, and that opens up a lot of possibilities. But SQL that runs is not always the same thing as an answer you’d confidently bring into a board meeting. Organizations still need certified metrics, governed relationships, and consistent definitions. Finance ARR and Sales ARR cannot quietly become different concepts depending on whether the answer came from a dashboard, an app, or an agent. Nobody wants three versions of ARR arguing with each other in a conference room. This is also where the distinction between a semantic layer and an ontology (or "business graph") matters. An ontology can provide a broader model of the business—its entities, concepts, processes, and relationships—while semantic views and layers provide the governed analytical definitions that tools and agents can actually use consistently. The future, to me, is not raw SQL generation versus a rigid compiler. It is an architecture that gives agents room to reason while grounding their answers in the same trusted semantics used by every other data experience. That combination of flexibility, determinism, *and* cross-client consistency is where this gets interesting. Great discussion, and kudos to the Bauplan team for bringing together the right people at the right time. https://lnkd.in/drmuEftvRethinking the Semantic Layer- Part II: The Builders ResponseRethinking the Semantic Layer- Part II: The Builders Response
-
Simba Khadder liked thisCoolSimba Khadder liked thisThe future of agentic AI won't be defined by better models, but by better context. Simba Khadder, our head of AI products, joined Dr. Alexander Borek on the Data Masterclass Podcast to discuss why context engineering, not prompt engineering or bigger foundation models, will determine who wins with AI in 2026. Their conversation traces the shift from RAG to context engineering and what it means for teams building agents that need the right data at the right moment, not just a bigger context window. Watch the full interview for a practical look at context engines, AI sovereignty, and what it actually takes to build production-ready agents: https://lnkd.in/es4Gi3wW
-
Simba Khadder liked thisSimba Khadder liked thisThe future of agentic AI won't be defined by better models, but by better context. Simba Khadder, our head of AI products, joined Dr. Alexander Borek on the Data Masterclass Podcast to discuss why context engineering, not prompt engineering or bigger foundation models, will determine who wins with AI in 2026. Their conversation traces the shift from RAG to context engineering and what it means for teams building agents that need the right data at the right moment, not just a bigger context window. Watch the full interview for a practical look at context engines, AI sovereignty, and what it actually takes to build production-ready agents: https://lnkd.in/es4Gi3wW
-
Simba Khadder liked thisSimba Khadder liked thisWhat most teams think is a data problem is actually a context problem. Customer data lives in one system and policies live in another. Product info, conversation history, and operational signals are all scattered across APIs, databases, and SaaS tools. The real bottleneck isn’t the model; it’s getting the right information at the right moment. That's the discipline behind context engineering: giving agents fresh data, relevant memory, trusted knowledge, and real-time signals instead of a single slice of the picture. It's also why RAG alone doesn't solve this anymore. Production agents need chunk-based retrieval for documents, MCP-style tools for structured data, and agentic RAG to decide what to fetch and when. Memory and semantic caching are part of the same system, and teams routinely mix them up. Memory remembers what's true about a user, account, or project. Semantic caching skips a redundant LLM call when a similar question has already been answered safely. One builds continuity. The other cuts cost and latency. Confusing them is how a cache ends up serving the wrong answer to the wrong account. Simba Khadder built out an FAQ that answers common questions about building better AI agents with real-time context, memory, RAG, and semantic caching. Full FAQ here: https://lnkd.in/eN72FNtX
-
Simba Khadder liked thisSimba Khadder liked thisIntroducing Lotte Verheyden 🇧🇪 Since the start of the year, we've doubled the team. We nearly missed to share our latest hires with you. We're making up for this now. Lotte joined us to build out our product marketing and developer relations efforts. She moved to San Francisco to be close to the action. A few things she's already shipped: 📚️ Launched Langfuse Academy, a hands-on guide to evaluating and improving AI apps, with 1M+ impressions and ~500 new followers in its first days. 👩💻 Built the Langfuse skill for AI coding agents, so anyone can add full observability just by asking their agent to set it up. 🌎️ Took Langfuse on the road with hands-on workshops at Open House and AI Engineer World's Fair and many Bay Area meetups. It's great to work with you, Lotte
-
Simba Khadder reacted on thisSimba Khadder reacted on thisIf you've tried setting up Hermes or OpenClaw, you quickly learn that designing memory/context is extraordinarily complicated. Cassidy Williams' blog mentions Redis as the "boring" infra answer to the context puzzle, reminding me of the terrific talk Simba Khadder gave at AI Council this year. His talk introduces "Context Engineering 2.0" - a systems-level approach to building agentic AI where context is treated as first-class infrastructure. Scalable agentic systems require a context engine. Useful resource: https://lnkd.in/eaxeNtZ3
Experience
Education
Projects
-
Slug Route
-
See projectBefore leaving UCSC, I released the bus tracking app that's still being used years later.
- In the news: www.santacruztechbeat.com/2015/06/18/qa-real-time-campus-shuttle-tracker-slug-route/ -
Other similar profiles
Explore more posts
-
Sushant Gupta
AI workloads are growing fast, so we’re engineering the full stack—from silicon to software to the datacenter. Maia 200 brings high-performance inference, efficient FP4/FP8 throughput, and better cost-to-serve across Azure’s fleet. Maia 200 is now running real AI workloads in Azure. This is systems-level innovation in action. Learn more in our blog: aka.ms/Maia200blog.
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content