Granica

Is your organization truly prepared to tackle the data demands of AI? Data readiness is the cornerstone of AI success, yet many enterprises struggle to bridge the gap between their data strategy and AI aspirations. Ensuring data is not just accessible but also safe, reliable, and free from harmful content has never been more critical. The truth is, data readiness can feel daunting - even haunting. But don't let it give you nightmares! In our latest October newsletter, we'll show you how Granica is helping customers make data readiness less scary, and cover a range of relevant industry news. Be sure to give it a read and face your AI data challenges head-on this Halloween season. Enjoy! ?? ??

The best way to reverse engineer the best course of action for your customer? Employ the Hearts, Heads, Hands framework with your team. By uniting individual opinions under the same larger purpose, this framework invites the best kind of collaboration between teammates - which translates into the best kind of execution for serving customers. Here’s how it works: ??HEARTS This is where you determine your mission and vision for what you want to do in the world. Everyone should be 100% aligned on this mission and vision so that when you move on to the next two steps in the framework, you’re operating under the assumption that everyone’s heart is in the right place. ??HEADS This is where you allow disagreement and diverse opinions to flow freely. You encourage people to butt heads for a purpose, bringing their unique perspectives and context to each conversation. And when they argue, they’re doing so for a purpose - they’re ensuring that each decision is the right one for the organization and the customer. ??HANDS:? This is execution mode, where you take the culmination of the hearts and heads steps and put it into practice, where it will actually affect customers. If the first two steps go well, the decisions you make at this stage will usually be right. This framework works because it establishes the necessary goodwill for ideas to be tested and argued over in a psychologically safe environment… and then it surfaces only the best ones. So, if you want to feel good about making the right decision for your customers, start by getting your heart in the right place. You’ll know where to go from there.

Are you automatically a great entrepreneur if RJ Lumba hosts you on the Great Entrepreneurs podcast? Asking for a friend ?? Thank you to RJ for giving me the opportunity to talk about the foundation that underpins all building and scaling of AI models: data. In the course of the conversation, we discuss that, although intelligence emerges from scaling both data and models, the bitter lesson is clear: compute and model scaling alone won’t cut it. We’re increasingly running into data walls, which now represent the primary bottleneck. For more about this and other challenges in our space, and their implications for the future of AI, listen to our conversation - linked in the comments.

We had an amazing time at last month's inaugural Granica Tech Talk where Zhe Zhang, Distinguished Engineer at NVIDIA, discussed how to best leverage Ray and other RPC approaches to power ML. ??? Great discussions, networking with brilliant minds, and of course, delicious pizza and drinks. ???? Swipe through the photos to catch the highlights! ?? A big thank you to everyone who joined us and made the event so special. Looking forward to more like this! #Networking #TechCommunity #GoodTimes

?? Just 1 day left! Join us for an exclusive tech talk with Saurabh Vishwas Joshi, ML Lead at Pinterest, sharing insights on building real-world recommender systems at scale. ?? Granica HQ, Mountain View ? Doors open at 6 PM ?? Great food, drinks, and company Grab your spot now—don’t miss it!

Granica is heading to Amazon Web Services (AWS) re:Invent in two weeks! ?? We’ll be showcasing our game-changing data compression solution for petabyte-scale data lakes. Join Rahul Ponnala, Tarang Vaish, Jason Nadeau and other Granica Ninjas at booth #168 to chat about your AI use cases and check out demos of the Granica AI Data Readiness platform—it controls costs AND speeds up data processing for Spark/EMR/Trino/Hive environments. How? By adaptively compressing data lake files in columnar formats like Parquet to deeply and losslessly reduce file sizes by up to 60%—even if data was already snappy compressed! ????? Book a meeting - and bring all your burning questions. ??? https://lnkd.in/gyxBUzqD ?#AWSreInvent #AWS_Partners #AIReadyData #CloudCostOptimization

?? Powering ML with Ray: Insights from Zhe Zhang's Tech Talk ??? Last month, we had the privilege of hosting Zhe Zhang (Distinguished Engineer at NVIDIA and Ray expert) at Granica HQ in Mountain View ?? Zhe delivered a “no-BS”, thought-provoking talk that dove into a critical question: ?? How do you command and utilize remote servers effectively for ML workloads? Here are the ?? takeaways from Zhe’s talk: 1?? RPC is foundational: The concept of Remote Procedure Calls (RPCs) is the building block for distributed applications, enabling computation across remote servers. 2?? The cluster spectrum: Frameworks like K8s, Ray, and Apache Spark allow us to scale workloads across clusters, each offering unique tradeoffs for ML practitioners. 3?? Choosing the right framework: - ?? K8s shines for workloads resembling American football (quarterback passing to multiple runners). - ? Ray excels when your workload mirrors soccer (everyone passing to everyone). - Apache Spark simplifies things—if your workload fits their expressive language, it "does the RPCs for you." We’re incredibly grateful to Zhe for sharing his practical, real-world perspective on ML infra—one shaped by his hands-on experience at NVIDIA and Anyscale. ?? The session was both insightful and approachable with a bunch of helpful Q&A, and his session is a must-watch for anyone navigating the challenges of scaling ML workloads ?? Thanks again, Zhe, for bringing your expertise to Granica HQ! ?? ?? Check out Zhe’s blog (with the session recording inside) using the link in the comments. #MachineLearning #Infrastructure #DistributedComputing #MLPractitioners

You can’t build a generational company when one person is making all the decisions. You need the collective wisdom and leadership of a group of talented people. Not only does this keep a CEO from experiencing decision fatigue, it ensures that one person’s myopic view doesn’t dominate the direction of the company. Most companies have an iceberg structure - where the CEO and top leaders are visible, and the people who do a lot of the heavy lifting to drive the company vision forward, are hidden beneath the waterline. I intentionally flipped this structure at Granica, because I feel strongly about empowering individual leaders to make decisions. My responsibility is to set the vision, hire the right people to execute it, and ensure they are equipped with what they need along the way. With a reverse iceberg in place, we’re able to draw on the intelligence of all our leaders - not just one time-strapped CEO’s. Or as the old saying goes - “If you want to go fast, go alone. If you want to go far, go together.”

?? The latest edition of The Data Foundation is here! ?? Packed with fresh insights on AI, data, and cost savings to fuel your 2025 plans. This edition includes: ?? Major wins for Granica in the AI space ?? Simple strategies to slash data lake costs ?? Exclusive tips from Gartner on AI readiness ?? How AI and data are shaping climate action Plus, top industry news you won't want to miss. Curious? Subscribe today!

?? Excited to announce Episode 2 of Granica Tech Talks, featuring Saurabh Vishwas Joshi, ML Engineering Lead at Pinterest, discussing the critical challenges and solutions in training recommender ML systems at scale. From your Pinterest pins to your Netflix queue, from Spotify playlist to TikTok's addictive feed - recommender systems shape our daily experiences. Yet building these ML systems at scale remains one of tech's most fascinating engineering challenges. Join us for an evening of deep technical insights where we'll explore: -Scaling challenges in recommender system training -Core efficiency bottlenecks -Distributed scaling with Ray -Infrastructure optimization -Data pipeline performance tuning If you're building ML systems at scale, this is your chance to cut through the hype and get actionable insights from Pinterest's production experience. ?? Location: Granica HQ, Mountain View ? Time: 6:00 PM - 8:00 PM on Wednesday, Nov 20th ?? Pizza and beer provided Limited seats available. RSVP here: https://lu.ma/g4ms4h87