Prashant Tiwari

It’s been really interesting watching how the whole idea of a Semantic Layer has gone from “No, we don’t need that” to “Surely AI can figure that out”, to “Wait… we definitely need this.” The organisation has basically travelled that whole arc over the last 18 months. It was a passionate bet myself, Jonathan Tierney and Michael Perratt started pushing nearly 18 months ago, so watching discussions evolve like they have, I must admit, has been tinged with a bit of pride... And now? We’re kicking off broad initiatives and having serious conversations internally and with leading vendors about what the Semantic Layer actually is, who owns which bits, and how it all fits together. Clue: Foundational Semantics and Semantic Signals are not the same thing - and understanding the nuances of that difference is where things really start to click. The most positive outcome is seeing broader realisation you need both if you want things like GenBI / Conversational Analytics to be reliable and actually useful at scale. AI isn’t some mystical creature that flawlessly just intuits your business logic. It needs the foundation in place: the solid, agreed data models and definitions - and then it can layer on the more dynamic stuff like signals: hints, context, intent, instructions, behaviour. That’s where interactions start to feel intuitive, adaptive, even personalised. And that’s why the conversations are suddenly getting very real: • Who actually owns the Foundational Semantics? • Where do Semantic Signals come from? • Who curates and trains the signals? • How do we iterate fast on signals without breaking the core definitions? • How do we serve foundations and signals through one governance layer regardless of the consuming tools? • And how on earth do we get a production-grade Semantic Layer working across teams that all speak slightly different and often validly different “semantic languages”? My view: the Semantic Layer is about to shift from being seen as a technical component to becoming an organisational capability. Something shared across data teams, analytics teams, domain experts, platform teams and the AI enabled stacks. It’s taken a bit of time to get here, but the shift seems obvious now. AI won’t replace the Semantic Layer - it stands on top of it. And the organisations that clock this early enough are the ones that will get real, scalable value from generative analytics solutions… not just cool demos. More to come as we keep shaping this, while closely watching other trends in innovation in this space, but the momentum is definitely growing…

47

6 Comments

🚀 AI Engineering is evolving. Most engineers know DevOps. Many understand MLOps. But now a new discipline is emerging: LLMOps. Let’s break down the difference between: • DevOps • MLOps • LLMOps in the simplest way. ⚙️ 1️⃣ DevOps DevOps focuses on software delivery pipelines. Typical workflow: Plan ⬇ Code ⬇ Build (CI) ⬇ Test ⬇ Release ⬇ Deploy ⬇ Monitor The goal is faster and reliable software delivery. Key principles: ✔ CI/CD automation ✔ Infrastructure as Code ✔ Collaboration between dev & ops ✔ Continuous monitoring DevOps powers modern platforms like microservices and cloud-native systems. 🤖 2️⃣ MLOps MLOps adapts DevOps principles for machine learning systems. The pipeline becomes model-centric. Typical stages: Data Collection ⬇ Data Preprocessing ⬇ Feature Engineering ⬇ Model Training ⬇ Model Evaluation ⬇ Model Registry ⬇ Deployment ⬇ Monitoring Monitoring includes: • Data drift • Model decay • Accuracy tracking • Continuous retraining MLOps ensures ML models remain accurate in production. 🧠 3️⃣ LLMOps LLMOps is the next evolution for Generative AI systems. Instead of training models from scratch, we manage foundation models and LLM workflows. Typical pipeline: Foundation Model Selection ⬇ Prompt Engineering ⬇ Context / RAG Setup ⬇ Fine-Tuning (optional) ⬇ Evaluation ⬇ Deployment ⬇ Monitoring Key monitoring metrics include: • Hallucination detection • Bias & toxicity checks • Token usage & cost • Human feedback loops LLMOps focuses on operating large language models safely and efficiently in production. ⚡ The Evolution of AI Infrastructure DevOps → Software pipelines ⬇ MLOps → Machine learning lifecycle ⬇ LLMOps → Generative AI & foundation model systems 🔥 Why this matters Modern AI platforms now combine all three layers: DevOps → Infrastructure automation MLOps → Model lifecycle management LLMOps → GenAI system orchestration This stack powers: • AI copilots • Autonomous agents • Enterprise AI search • AI-native applications 💬 Quick question for AI engineers Which pipeline are you currently working with? 1️⃣ DevOps 2️⃣ MLOps 3️⃣ LLMOps 4️⃣ Hybrid AI stack Comment below 👇 #ArtificialIntelligence #GenerativeAI #DevOps #MLOps #LLMOps #AIEngineering #MachineLearning #CloudComputing #AIInfrastructure #DataEngineering #AIAgents #RAG #GenAI #TechInnovation #DeepLearning #MLOpsEngineering #TechCommunity #FutureOfAI 🚀

33

I'm genuinely humbled to share that 5-Spot has been accepted into FINOS. FINOS, the Fintech Open Source Foundation, part of the The Linux Foundation, just merged our project contribution PR into their Technical Oversight Committee repo. That's not something I take lightly. 5-Spot is an on-prem, CAPI-based spot computing scheduler for underutilized physical hosts. The idea is simple: financial institutions have enormous amounts of donated or idle bare-metal compute sitting in data centers doing very little. 5-Spot lets platform teams reclaim that capacity as interruptible, Kubernetes-native workload targets — without the cloud bill. This started as a problem I kept running into at RBC Capital Markets. We had physical servers that were graciously donated by internal teams — not cloud VMs, not leased capacity. Real servers, woefully underutilized. I wanted to build something the broader financial services industry could use and contribute to, not just something we kept behind a firewall. So I submitted it to FINOS. Having the TOC review, discuss, and accept 5-Spot into the foundation means this work now belongs to the community — not to me, not to any one firm. That's the part I find most meaningful. If you're working on underutilized on-prem compute, platform engineering in regulated environments, or CAPI-based scheduling, I'd love to connect. This is just the beginning. https://lnkd.in/e955yG4C #OpenSource #FINOS #Kubernetes #PlatformEngineering #FinTech #CAPI #SpotComputing #FinancialServices

61

5 Comments

**Bridging Borders, Expanding Opportunities: The UK-India Trade Deal in Focus** In an era where global collaboration drives economic progress, the new **UK-India trade agreement** marks a significant milestone. With **tariff reductions across key industries**, this deal not only strengthens bilateral trade but also opens new avenues for professionals and businesses alike. 🚗 **Automotive**: A game-changer—UK luxury cars will now be **far more accessible** in the Indian market. 🥃 **Whisky & Spirits**: With reduced tariffs, British distillers have a golden opportunity to expand their footprint in India’s thriving beverage industry. 💻 **Tech & Services**: Indian professionals will find **easier mobility**, fostering innovation and cross-border expertise. As we navigate this dynamic shift, **leaders and professionals must adapt**—seizing opportunities, forging partnerships, and amplifying their global presence. 💡 What are your thoughts on this agreement? How can industry leaders leverage this trade transformation? Let’s discuss! #Leadership #GlobalTrade #UKIndiaTrade #Networking #BusinessGrowth

12

1 Comment

MCP - a open protocol for AI (LLM) to talk to other tools / systems / applications. Naive, but don’t judge me, I was judging AI capabilities with me being Naive   - So I asked Claude AI to create an MCP - An MCP the AI itself will use - Gave a website address to which MCP had to be created (with public APIs) - It Generated extensive, detailed code - I dwelled on it, fine tuned for my preference / requirements, it went ahead and did a bit more than I asked for. Website was analysed in minutes (amusing and scary) - I asked it to package it as executable package - I had configured file system MCP - So, Claude leveraged that and created a python project in filesystem and created Project with proper files and folder structure - Here comes Claude code (CC) - This python project of MCP server, was not in a running state - Asked CC to make the code runnable (of course I was in control (manual loop) of the changes it was making) - MCP server to run in local system (MacOs) needed some changes in trivial python code, CC made the changes - Interesting side part, during this whole process, it was also creating documentation, to install server, to run server and the MCP config for other tools - At times I had to say update the documentation - OOPS !! Code got changed, but what was changed ? - I asked CC to make this project version controlled. Oh my Devaa (God), it converted the project into a Git repo (local) - Created git related files, Auto committed all files. All I had to do was, say yes, go ahead - Well, we got side tracked, lets get back to MCP - So, now I asked CC to run the server (lazy to copy the command and run it in next tab terminal) - Obviously.. It had issues, and obviously it auto corrected the code - It wrote test cases, wrote test start script, tested it, committed the code and then - now it tells me, 'All fixed, Want to run?', Howdhu maga howdhu, run (yes dude yes, run) - MCP server started. I updated MCP configuration and restarted Claude Desktop - pssssttt….. curios? - :( Error: MCP server couldn’t be started - Again me being extremely (purposefully) lazy, told CC, you are being dumb, fix the startup error - I guess, I hurt its Ego. - It asked permissions for MCP server log file. whaaaat really ?, it wants to read log files now, wow. - It test ran with its test scripts, read the log, analysed like a nerd, updated the test files and code, reran until it was fixed and updated the documentation and also committed the code - What can I say, I was watching Christopher Nolan movie level events. - Finally, MCP worked, I tested few things, results were as expected.   It ain’t over, my architectural / Engineering hormones kicked in. How about containerisation ? How about productionising it ? How about Security & Authentication ? How about Performance & Scalability ? How about Operational Readiness ? How about AI Security, Gaurdrails ? Stay tuned, I will be back 🙂

14

LRU ( Least Recently Used) cache is one of widely used data-structure in memory management Few examples: Spark memory management: when you cache a data frame using cache() or persist(), it stores blocks of data in memory but memory is limited based type of compute cluster selected in cloud. so, when its full, spark uses LRU to evict blocks of least recently used data to make space for new data Embeddings or Vector cache (in RAG): Embeddings of past used queries or documents are cached. LRU is used to evict old embeddings from in-memory stores or vector caches when new ones are added. I just tried to implement simple LRU in python myself. Its really worth it :) Real world LRU's are definitely complex, give it try if you are interested.

16

Ever wondered how modern chat apps deliver messages in the blink of an eye? The secret sauce is often Redis Pub/Sub, a lightweight, high-performance messaging system that powers real-time communication. Let’s break it down! 🔑 What is Redis Pub/Sub?     Redis Pub/Sub is a publish/subscribe messaging model where publishers send messages to channels, and subscribers receive them in real time. No direct contact between clients is needed, and messages are delivered in under 10ms on average—faster than your reaction time! ⚡ It’s perfect for chat apps, notifications, or any system needing instant updates. 🛠 How it Powers Your Favorite Chat Apps: Channels: Each chat room or conversation is a Redis channel.   Publishers: When you send a message, the server publishes it to the channel.   Subscribers: Users in the chat room subscribe to the channel and get messages instantly.   Message History: Redis Pub/Sub doesn’t store messages, but you can pair it with Redis lists or a database for chat history.   Disconnects: Users can unsubscribe when they leave, keeping things efficient. 🌟 Why It’s Awesome   Speed: Near-instant message delivery for seamless user experiences.   Simplicity: No complex setup—just subscribe, publish, and go!   Versatility: Beyond chat, it’s used for notifications, job queues, analytics, and more. 🧠 Beyond Chat: Versatile Use Cases! Redis Pub/Sub isn't just for chat. It's a versatile tool for: Real-Time Updates: Instant news feeds, live score updates, or dynamic dashboard refreshes. Notifications: Sending immediate alerts for order status, product availability, or system events. Distributed Systems Coordination: Enabling seamless communication and event triggering between different services. Job Queues: A simple yet effective way for worker nodes to pick up and process tasks. Analytics: Streaming events for real-time data processing and insights. 💡 ENGAGE WITH ME !  : We recently put these capabilities to the test! We performed extensive load testing and performance testing using JMeter to simulate real-world scenarios. Our tests involved both publishing and subscribing to messages from Redis Pub/Sub channels and Redis Streams. To gain deeper insights into the system's behavior, we leveraged Redis Key space  notifications to monitor key creation, expiration, and deletion statuses in real-time.         Have you used Redis Pub/Sub in your projects? What’s your go-to tool for real-time messaging—Redis, Kafka, or something else? Drop your thoughts below—I’d LOVE to hear your experiences or use cases! 👇 #Redis #PubSub #RealTimeMessaging #TechTalk

1

More or 'better' data is not going to cure LLM or DL problems.. For years, ML and AI projects have had this convenient excuse.. bad data or not enough data. It has been the escape hatch used to justify the staggering failure rate across the industry, with over 80 percent of projects falling short, as documented by research firms like McKinsey and Gartner. Yes it is true that the data is quite often bad or not enough or full of bias. But here is the thing no one wants to say out loud.. There may never be enough data. Ever(!). To make deep learning work in many mission critical open world contexts that require adaptive autonomy.. No matter how much we collect. No matter how well we clean it.. LLMs represent the first serious effort to scale deep learning in this autoregressive form to unimaginable magnitude. Trillions of tokens, hundreds of billions of parameters, nearly every scrap of digitized human output. And yet the models seem to perform worse and worse. Error rates have not improved. Hallucinations persist. In tasks that require reasoning or multistep inference, the latest models have sometimes even regressed. This is not speculation.. this is confirmed in ClosedAI's own published benchmark results. Despite having access to nearly all known data, including stolen IP, and astronomical compute, the outcomes are plateauing or degrading.. not improving. The hard truth is this. Even if we had every piece of data on Earth, multiplied by a factor of a thousand, and increased the dimensionality of the embeddings to absurd levels, it still would not be enough. Size here does not matter.. Neither data size nor model size nor hidden dimensionality guarantees coherence, accuracy or robustness. With LLM hyperscaling of all available human data, what we achieved was density. but not uniformity. And due to the very nature of LLM architecture, any sequence drifts. Each generation is a stochastic walk. If the early tokens fall into low probability basins, the entire chain follows that trajectory. The drift is not an accident. It is built into the process. It is structural. The longer the sequence, the higher the risk of divergence. Every output is a compounding approximation with no global correction mechanism. Synthetic data does not solve this either. Attempts to patch coherence using synthetic reinforcement only accelerate degradation.. Soon enough, it should become apparent to everyone. But it seems that each new generation of researchers and executives forgets what has already been learned. Institutionalized knowledge is ignored. Hype cycles repeat. The same core issues are reframed as if they are new with new anthropomorphic terms like 'hallucinations.' The limitations of deep learning are not just data related. They are systemic. They are foundational. And data will not always be the convenient scapegoat for failure.. That much is clear. #ai

98

37 Comments

📐 Every distributed system must make an impossible choice. The CAP Theorem says: You can only guarantee 2 out of 3 👇 🔵 Consistency — Every read gets the most recent write 🟢 Availability — Every request gets a response (even if outdated) 🔴 Partition Tolerance — System works even if nodes can't communicate How big companies chose: 🏦 Banks → CP (Consistency + Partition) — wrong balance = disaster 🛒 Amazon Cart → AP (Availability + Partition) — better to show stale data than crash 🔍 Google Search → AP — availability wins for user experience In reality, network partitions ALWAYS happen. So you're really choosing between Consistency vs Availability. 💡 Ask yourself: "What's worse — stale data or no data?" That's your answer. ♻️ Repost if this clicked for you! Follow for more system design concepts explained simply. #SystemDesign #DistributedSystems #SoftwareArchitecture #Engineering

13

Pretty cool Kunal Kishor - not that I understood it..but if a GCP technologist like you have written - it must be true! Nice post TANMOY KANUNGO Rajdeep Banerjee

Building a trading platform that holds up during market surges has been a key engineering focus at INDmoney. AWS recently published a case study on the INDstocks trading engine, outlining how the platform handles millisecond-level order processing and real-time price delivery, even during peak loads. Good to see the team’s work recognised. AWS Case Study: https://lnkd.in/gFsN_xpd Phenomenal team work by Deepak Pandey Saksham Jain Sonu Singh Kapil Ramwani Tushar Kumar Singh Himanshu Agrawal Kartikay Aggarwal Vinayak Gupta Meenakshi Sundaram M.

69

1 Comment

CAP Theorem — continuing from the previous post In the last post, we looked at what breaks when a network partition happens. Now let’s look at the decision a system is forced to make. When nodes can’t talk to each other, there’s no neutral option. You either: • Stay Available (AP) Keep serving requests, accept temporary data divergence. • Stay Consistent (CP) Protect correctness, reject or delay some requests. Partition tolerance isn’t optional. Networks fail. So the real design question becomes: Which failure mode can your system tolerate when the network misbehaves? Good systems don’t avoid the trade-off. They design for it — intentionally. #DistributedSystems #CAPTheorem #SystemDesign #SoftwareArchitecture #BackendEngineering #Scalability #Engineering

12

💸 Automating EBS Snapshot Cleanup with AWS Lambda - Aligning with the Cost Optimization Pillar of the AWS Well-Architected Framework In many AWS environments, EBS snapshots accumulate quietly over time-from terminated instances, outdated backups, or forgotten test environments. Each of these snapshots incurs ongoing storage costs, often hidden until you review your bill in detail. To address this, I created a simple AWS Lambda function using Python and Boto3 that automatically detects and deletes unused EBS snapshots, helping keep your environment clean and cost-efficient. 🚀 🧩 How it Works The Lambda function: Connects to EC2 via Boto3. Fetches all EBS snapshots owned by the account. Lists all running EC2 instances. Deletes any snapshot that: Has no associated volume, or Belongs to a deleted or detached volume. ⚙️ Why AWS Lambda? Serverless - No infrastructure to manage. Event-driven - Can be scheduled using CloudWatch Events (EventBridge) to run daily or weekly. Cost-efficient - You only pay for execution time (usually a few seconds). 🔐 Required IAM Permissions {  "Effect": "Allow",  "Action": [   "ec2:DescribeSnapshots",   "ec2:DescribeInstances",   "ec2:DescribeVolumes",   "ec2:DeleteSnapshot"  ],  "Resource": "*" } 📊 AWS Well-Architected Framework - Cost Optimization Pillar This automation directly supports the Cost Optimization pillar of the AWS Well-Architected Framework by: Identifying and removing unused resources (EBS snapshots). Implementing cost control mechanisms through automation. Measuring and optimizing ongoing usage to avoid waste. 💡 In other words, it’s a practical “quick win” that enforces cost hygiene through automation. 🧠 Bonus Tip You can easily extend this function to: Implement a “Dry Run” mode to simulate deletions. Integrate with Slack or SNS for notifications. Use AWS Config or Trusted Advisor to detect snapshot sprawl before cleanup. 🔗 If you’re working on AWS cost optimization or Well-Architected reviews, automating snapshot management is one of the easiest and most impactful steps to take. Here is Github repo link: https://lnkd.in/etCQFuKQ #AWS #Lambda #Boto3 #EBS #CloudCostOptimization #WellArchitectedFramework #Serverless #Automation #DevOps #Python

20

2 Comments

Interesting topic came during a discussion - Can dlt from dltHub be used with Plotly and I said why not. Pythonic developers would love it. Here are the suitable candidates; I think 1) Real-Time API Data Dashboards Utilize dlt to extract data from APIs and load it into a database like DuckDB or MotherDuck. Then, use Plotly to create interactive dashboards that visualize this data in real-time.​ 2) Data Quality Monitoring Implement data quality checks within your dlt pipeline and visualize the results using Plotly. This helps in identifying anomalies or data issues promptly. 3) Schema Evolution Tracking Monitor changes in your data schema over time using dlt's schema evolution features and visualize these changes with Plotly. 4) Business KPI Dashboards After transforming and loading your business data using dlt, create dashboards to monitor key performance indicators (KPIs) with Plotly. 5) Interactive Data Exploration with Dash Combine dlt with Plotly Dash to build interactive web applications for data exploration. This is an interesting topic which I would like to cover in future. Another workshop session, who knows?

17

2 Comments

Will Tokenisation Unlock Many Locked Assets? Is Digital Transformation Underway Through Asset Tokenisation? According to Boston Consulting Group (BCG), the market opportunity for asset tokenisation is projected to reach a staggering $16 trillion by 2030. In simple terms, tokenisation refers to the digital representation of ownership rights to an asset using tokens on a blockchain. This innovation has the potential to revolutionize how we hold and transfer assets from financial instruments like certificates of deposit to real estate to art. Imagine owning a fraction of a commercial building - Asset tokenisation enables fractional ownership in small units/small division, making high-value assets more accessible and liquid. 🔍 Key Benefits of Tokenisation: 💧 Enhanced liquidity 🔁 Seamless transfer of ownership 📉 Fractional ownership of high-value assets 🌍 Global accessibility 🔍 Greater transparency via blockchain ⚡ Real-time transactions ⚖️ Automated compliance, dividend distribution, and governance Of course, few challenges remain such as legal recognition, acceptance, security risks, token management etc., are areas that need attention. Encouragingly, the Reserve Bank of India (RBI) has launched a pilot for tokenised digital assets, starting with Certificates of Deposit. This initiative is expected to unlock new markets and enable faster, safer, and more cost effective transactions. As per recent fintech research, 63% of custody banks already offer tokenised assets, with another 30% planning to adopt within the next two years. The momentum is building. It’s the right time for financial institutions, banks, fintechs, and tech companies to embrace this billion-dollar opportunity. #AssetTokenisation #DigitalAssets #Fungibility #DigitalAdoption https://lnkd.in/dyRzJwk2 https://lnkd.in/dEu_gZzc

9

Hello new maven central. (Painless Migration) Goodbye old build pipeline Chaos. It's done. All my repositories have now ascended to Maven Central. Yes, that Maven Central. The one most people fear migrating to, as if it required a blood sacrifice and a signed letter from your DevOps overlord. Spoiler: It doesn't. Turns out, the process is quite… tolerable. Even elegant. With the new generic deployment pipeline here, I now queue and release everything directly from Maven Central's side. No more babysitting release builds in CI. No more elaborate tribal rituals involving thousands of YAML files and a sacrifice to the cron gods. It's scheduled, structured, and blissfully boring. ``` on:  schedule:   - cron: '0 4 * * *'  workflow_dispatch: jobs:  build:   uses: YunaBraska/YunaBraska/.github/workflows/wc_java_build.yml@main   with:    run_update: true    run_test: true    run_deploy: false   secrets: inherit ``` Everything is unified. Even the build files. Even the versioning, now writing clean date-based semver like 2025.06.XXX. It looks better. Feels cleaner. Like watching a garbage fire finally extinguish. Only real pain? GitHub personal repos. No org = no shared secrets. I had to inject secrets one-by-one, repo-by-repo, like some kind of YAML nurse. Disgusting. But here's what really saved me during deployment: Tests. Real tests. No mocks. No Mockito, no fake behaviors. Just real, running components validating real scenarios. My test suite knows if the thing is broken. No imagination required. Because I avoid unnecessary dependencies like most people avoid responsibility, the tests run fast. Blazing fast. Jet-engine-on-fire fast. They are the documentation. The security. The QA team I never hired. These pipelines update the repo automatically since years! Not by humans or AI. No Dependabot vomiting PRs into my inbox. Just logic that finds the right Java version, detects the build tool, and does the job every time. The full migration? Took half a day. In corporate terms, that's less than a single "let's align expectations" meeting. Compared to the enterprises I've worked in, where such a move would take years and require three full-time Slack channels, this was laughably smooth. Final thought: OSS is the infrastructure of the internet. Most companies don't pay for it. They just consume. If maintaining it was counted as work time, the world would break less. But hey, if I can migrate everything to maven central in a few hours, maybe you can too. Unless, of course, your pipeline still cries when you upgrade Java without dependabot or has a too high complexity. In that case… good luck. Links for the curious: 📦 Reusable GitHub Workflow: https://lnkd.in/dS7-P5gs ⚙️ Example Run: https://lnkd.in/dAypTfrD Deploy safely. Or don't. I won't stop you.

30

3 Comments

⚡ Microservices Optimization with Apache Kafka As microservices architectures scale, one challenge becomes clear: 👉 How do we ensure efficiency, reliability, and seamless communication across distributed services? This is where Apache Kafka plays a transformative role. 🔹 Event-Driven Backbone – Kafka decouples services, enabling real-time event streaming and asynchronous communication. 🔹 Performance Optimization – Reduces point-to-point integrations, lowering latency and improving throughput. 🔹 Resilience – With replication and partitioning, Kafka ensures fault tolerance and high availability. 🔹 Scalability – Horizontal scaling of both producers and consumers allows systems to grow without bottlenecks. 🔹 Observability – Metrics, lag monitoring, and alerting integrated into CI/CD pipelines keep the system transparent and reliable. In my experience, combining microservices optimization techniques (caching, async processing, bulkheads, circuit breakers) with Kafka as the event backbone creates a platform that is not just scalable — but also resilient and future-ready. #Microservices #Kafka #Architecture #Scalability #Resilience #EngineeringExcellence

7

I just published a new blog on Elasticsearch — explained as a story We often talk about Elasticsearch in terms of shards, inverted indexes, and latency numbers. But why does it actually work so well at scale? In this post, I explain Elasticsearch as a living city: A librarian who never scans every book (inverted index) A city split into districts (shards & replicas) Queries answered in milliseconds, even with billions of documents Real performance numbers, not just theory If you work with search, logs, analytics, Kafka pipelines, or distributed systems, this one’s for you. Read the full story here: https://lnkd.in/dMZUpAMW Would love to hear your thoughts or how you’re using Elasticsearch in production. #elk #elasticsearch

8