Articles by Arvind
Activity
5K followers
-
Arvind Prabhakar shared thisA few of my thoughts on Daniel's post below. Tabsdata materializes the full dataset before writing to destination. This is a prerequisite for correctness: joins, aggregations, transactional consistency, data quality, lineage, replay. It also happens to be exactly how modern data platforms like Snowflake, Databricks, AWS and more prefer to receive data: large Parquet files, bulk transfers, columnar writes. That architectural alignment is why the performance numbers look the way they do. The benchmarks are also honest about where Airbyte holds its own. What is not covered here: incremental loads, CDC, transformations. More to come.Arvind Prabhakar shared thisThe data is in! Tabsdata is up to 86x faster than Airbyte! 🚀 Over the past several weeks, I've been working on a really exciting benchmarking project comparing Tabsdata performance to other popular data integration tools. The first leg of this journey explores how Tabsdata and Airbyte compare when extracting and loading full data refreshes across different sources and destinations. Although extract and load is only a small subset of what Tabsdata can do, I wanted to see how it performs when restricted to point-to-point data movement, workflows that Airbyte specializes in. As the benchmarks show, Tabsdata vastly outperforms Airbyte, while remaining lightweight, offering complex data transformation capabilities, and providing automatic data versioning, lineage, and observability out of the box. There's still a lot more to explore: incremental CDC, ETL, testing more sources and destinations, and more. I am also creating Terraform scripts so anyone can run and validate these benchmarks. Read all about it in my blog post. https://lnkd.in/guHKDQTY
-
Arvind Prabhakar shared thisThe best data conversations happen over a cold one. If you're in the Portland area, don't miss this - good people and real talk about data engineering. RSVP and join Daniel and the team!Arvind Prabhakar shared thisHi everyone! For folks in the Portland area, I wanted to announce the second PDX Data Engineering Happy Hour which I'll be hosting at Binary Brewing (Beaverton Location) on March 19th @ 6:30 PM. Similar the our last event, it'll be informal and we'll have a free bar tab (courtesy of Tabsdata). Come by, grab some drinks or food on us, and chat with other people in the Portland data community. If you're interested, only requirement is just to RSVP through the link below https://luma.com/c133sajuPDX Data Engineering Happy Hour (March Edition) · LumaPDX Data Engineering Happy Hour (March Edition) · Luma
-
Arvind Prabhakar posted thisData is corrected after it moves. Categories are redefined and records are restated long after pipelines have done their job. Downstream systems keep operating on assumptions that are no longer true. I call this truth-drift.
-
Arvind Prabhakar posted thisThe modern data stack is really a compensation stack for pipeline-centric design. Most of our time and tooling go into rebuilding trust after data already moved. We normalized the lie that reliable transport is enough. It isn’t.
-
Arvind Prabhakar posted thisETL systems fail in more ways than people admit. Gaps between jobs, failures that force reruns, and shifts in execution order all create phases where data is updated but not logically consistent. Anyone who runs a data platform has seen this. Tucu (Alejandro Abdelnur) just wrote a clear breakdown of why this happens in modern stacks. He walks through execution plans, transactional boundaries across domains, and what it takes for downstream systems to see one coherent state. It is a short practical read and the first article in our new Tabsdata publication. Worth a look if you want cleaner refresh cycles and fewer surprises in your platform. More info in comments.
-
Arvind Prabhakar shared thisELT vs ETL: Most teams today follow the ELT model, where data is first loaded into a platform and then transformed inside it. ELT replaced ETL largely because transformations became impractical outside data platforms as data sources multiplied. Joining and shaping data across APIs, files, and other stores required complex logic for boundaries, retires, and failure handling. By shifting transformations downstream, teams simplified ingestion. ETL originally performed these steps inline, pushing SQL down to the source systems for efficient and consistent results. The trade-off has become clear. ELT fragments ownership and inflates cost. Ingest tools, transformation orchestrators, observability layers, and semantic modeling layers all exist to compensate for the loss of context between extraction and transformation. The outcome is slower refresh cycles, inconsistent metrics, and rising compute and storage expenses. Each layer adds friction, erodes accountability, and weakens trust in data. What started as a practical workaround has hardened into an architecture that drains resources and delays decisions. Tabsdata restores balance by bringing ETL back to its architectural roots. It unifies extraction, transformation, and load in a single declarative flow that scales horizontally across serverless infrastructure. Transformation occur as data is collected and results propagated to any destination. This reduces latency, cost, and operational complexity while preserving the shared context between data producers and consumers. It is the simplicity of classic ETL, now built for the modern data stack.
-
Arvind Prabhakar shared thisIt is common for different teams in the same organization to report different values for the same metric. Ask Finance and Operations for "net revenue by region last quarter," and you may get two different numbers. Add an AI assistant to the mix and you might get a third. Each team believes it is using the same data, yet the results diverge. This inconsistency erodes confidence and makes it difficult to align on business performance and next steps. The divergence rarely comes from data itself. It arises from how data is accessed and interpreted. Business terms lose precision because of misaligned semantics between domains, and multiple copies of data obscure the truth, blindsiding queries that would otherwise be correct. Metadata dilution strips away key details, introducing inconsistencies that accumulate over time. With uneven update timings, lack of dependency management, and unsynchronized copies, what started as a single source of truth fractures into several partial truths. Tabsdata prevents this by ensuring every dataset refresh propagates instantly and consistently across all dependencies in the correct idempotent order. Semantics and metadata remain intact, preserving a shared meaning of each dataset and all downstream results. And when differences do appear, Tabsdata's built-in lineage shows exactly where and why they occurred. The result is a foundation of trust and alignment that allows Finance, Operations, and every other domain to move with clarity and speed. Cross-functional consistency becomes the hallmark of agile, high-efficiency business.
-
Arvind Prabhakar shared thisStreaming and batch processing solve different problems. Streaming handles ordered, unbounded events that flow between applications. Batch handles unordered, bounded datasets designed for completeness and reproducibility. Streaming fits application integration and event-driven systems. Batch fits data integration or ETL. They are not interchangeable, and forcing one to act as the other creates unnecessary complexity. Many enterprises use streaming tools such as Kafka or Flink for data integration, assuming continuous flow will improve freshness. In reality, streaming semantics depend on delivery order, windowing, and timing assumptions that complicate analytics and require bespoke protocols between disconnected systems. What starts as a shortcut to speed often turns into an operational burden. The right way to gain speed in data integration is through instant ETL, where each dataset refresh propagates instantly and consistently. Tabsdata brings Pub/Sub semantics to datasets and delivers instant ETL for those who need speed without compromising consistency or trust. Streaming belongs where order defines correctness. Batch belongs where completeness defines trust. Mixing them serves neither well.
-
Arvind Prabhakar shared thisIt was great to be at TechCrunch this week and to meet so many teams who stopped by our booth to learn more. Most conversations centered on one concern: the rising cost of managing data across too many tools. Ingestion, orchestration, observability, data quality, semantic, and metadata tools each come with their own licenses and consumption models. The more the utilization, the faster the costs grow. Many teams turn to open source frameworks to control costs and stay flexible. These are powerful foundations, but as systems expand, maintenance and evolving best practices require dedicated expertise. The effort to manage change, dependencies, and failures often outweighs the savings. Add in wasted compute, redundant storage, unreliable governance, and the true cost becomes clear. The highest cost appears when data is not reproducible. Insights lose credibility, compliance becomes uncertain, and decisions drift from fact. Tabsdata brings all data operations - from ingestion to semantic cohesion - into one integrated platform. By removing redundant systems and operational overhead, Tabsdata delivers instant ETL and ensures every dataset is consistent, reproducible, and ready to use.
Arvind Prabhakar commented on a post
2mo
Thanks, Paul Dudley I really enjoyed the conversation. Honest exchanges like that, sharing what's worked and what hasn't, are how we actually learn.
Arvind Prabhakar replied to a comment
3mo
Prakriti, the legacy is the layered ELT pipelines in use today. The real friction comes from tight coupling across ingestion, staging, and transformation.
Tabsdata removes the coupling by unifying extraction and transformation in one flow. You migrate a slice in parallel, validate it, and cut over cleanly.. Each component migrated slice (datasets / transformation / derived datasets) becomes reusable, so later migrations move faster. The velocity builds as more of the stack shifts over, with better refresh cycles and consistency along the way.
-
Arvind Prabhakar reacted on thisArvind Prabhakar reacted on thisWhen Max Schireson joined Battery Ventures over a decade ago, the plan was for him to come in as an EIR and start another killer database company. I had already seen him grow MongoDB from 0 to tens of millions while I was on the board there, so everyone expected round two. Fortunately for us, he took a liking to the investing side and it's a great honor to officially name him as partner! Over the last 10+ years, I've watched Max sit with the founder of Databricks and go deep on open source metrics, then pivot to debating astrophysics and nuclear energy with a room full of engineers - and everyone walks away feeling like they were talking to one of their own. He's got this amazing versatile and broad spectrum paired with real operator instincts for helping companies navigate financing, growth, which a lot of people really enjoy, especially the founders who work with Max. As AI opens up massive oppty's in deep tech - foundation models, robotics, quantum, neuromorphic computing - Max has found an amazing cohort of founders - Fundamental, Quantum Art, Reflection AI and others building world-class foundational platforms in these spaces. We're excited to have him focused on these deeper tech ideas that we believe will potentially be game changers over the next 5-10 years!
-
Arvind Prabhakar reacted on thisArvind Prabhakar reacted on thisA little over a month ago, I wrapped up five years at Snowflake. What a five years it was. I had the privilege of building and leading the Customer Experience Engineering (CXE) team, surrounded by some of the brightest, most driven people I've ever worked with. Over the past year alone, the CXE team fundamentally reimagined how Snowflake's CX organization operates with AI, delivering substantial gains in customer self-service and efficiency. I couldn't be more proud of what we built together. But every great chapter eventually leads to a new one. Back in late February, I took the leap and began co-founding a company alongside my longtime friend and colleague Linden Hillenbrand. In the first week of March, we officially incorporated as CustOS AI (pronounced 'KOOS-tohs'). Our mission: helping enterprises transform their Customer Operations with AI and data. It's an intersection I've spent the last 15 years hyper-focused on within Cloudera and then Snowflake. I’m ready to bring this to the broader market. Since day one, we've been heads-down meeting with leaders across our networks, connecting with potential design partners, pressure-testing our thesis, and sharpening our thinking with every conversation. The energy has been incredible. If you’re interested in learning more, please don’t hesitate to reach out. While I know this journey won’t be easy, I genuinely believe we're working on a problem that can drive real impact for the market. This is just the beginning. More to come. 🚀
-
Arvind Prabhakar liked thisArvind Prabhakar liked thisManufacturing is entering its most important decade in 50 years. And AI will determine who leads it. Today marks one year since I became the CEO of QAD | Redzone. When I joined, I had a simple belief: Manufacturing doesn’t need more software. It needs systems that act. Factories today are navigating unprecedented complexity — tariffs, labor shortages, supply chain volatility, and margin pressure. And now AI is reshaping how decisions get made. The old model — systems that simply record what happened — is no longer enough. Manufacturers need platforms that help them sense faster, decide faster, and act faster. That belief has shaped everything we’ve done this past year. In 12 months we have: • Launched Champion AI — an agentic AI platform purpose-built for manufacturing • Unified Adaptive ERP and Redzone Connected Workforce around a Systems of Action strategy • Accelerated our pace of execution across the company But the biggest shift this year hasn’t been technology. It’s been clarity. Manufacturing is entering a defining decade. The winners will not be the companies with the most dashboards. They will be the companies that can turn insight into action the fastest. That’s the future we are building. To our employees, customers, and partners — thank you for believing in this journey and for all the support you have given me. Year One was about focus. Year Two is about acceleration. Manufacturing is at a once-in-a-generation inflection point. And we intend to lead it. #Manufacturing #AI #SystemsOfAction #ChampionAI #Leadership Bryan Reimer John Dyck Jeff Winter Jake Hall Chris Luecke Allison Roberts Grealis Matthew Littlefield Michael Rowe Joe Sullivan Yvonne Genovese mikeroweWORKS Eric Kimberling Mark Vigoroso, MBA "R "Ray" Wang Holger Mueller Patrick Moorhead Robert Kramer National Association of Manufacturers - NAM CESMII #MAPI
-
Arvind Prabhakar liked thisArvind Prabhakar liked thisI’m incredibly honored to be selected for the HITEC 2026 Emerging Executive Program (EEP)! Looking forward to a year of growth alongside 100 top-tier technology leaders from 70 organizations. A huge thank you to my team at Capital One for supporting the nomination and growth in my leadership journey! #HITEC #EEP2026 #TechLeadership #ExecutiveDevelopment
-
Arvind Prabhakar liked thisCome for the context engineering, stay for the unexpectedly strong opinions about the Oshawa Generals.Arvind Prabhakar liked thisWe’re proud to be at #CDAOCanada 2026, joining Canada’s leading data and analytics executives to explore how data, AI, and governance are driving innovation and smarter decision-making. Catch our speaker, Dima Spivak and be part of the conversation. Learn more: https://ibm.co/6043EDZBX
-
Arvind Prabhakar reacted on thisArvind Prabhakar reacted on thisMy Career Is My Mom's Cup of Tea On this Women's Day, I find myself without the most important woman in my life. My mother left us on January 21st of this year, and the world has felt differently weighted ever since. Tomorrow, as the world honors women everywhere, I want to honor just one — quietly, personally, with the full weight of everything she meant to me. In 1991, I was fighting for an engineering seat against some of the most brutal competition India has to offer. I was never the most talented in the room — but I was relentless, and she made sure I stayed that way. Night after night, when I would push through until the early hours, she would appear at exactly the right moment - just when I was about to fall asleep — not because of an alarm, but because she simply knew — with a cup of tea, with her presence, with words that kept me going When I finally fell asleep, she would quietly put my books in order, as if preparing the battlefield for the next day's fight. That was the kind of love she gave — precise, selfless, and always on time. The career that followed was built on a foundation she laid in those quiet pre-dawn hours. Every moment I have chosen service over comfort — that was her. I was on a flight to say goodbye on January 21st, and I didn't make it in time. But I've come to understand that with a love like hers, the goodbye was never really the point. The point was every cup of tea. Every book she placed back in order. Every morning, she made it possible. Happy Women's Day, Maa. The most important woman in my life — then, now, and always. Your's always - Raju
Experience & Education
-
Tabsdata
View Arvind’s full experience
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
Other similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top contentOthers named Arvind Prabhakar in United States
4 others named Arvind Prabhakar in United States are on LinkedIn
See others named Arvind Prabhakar