Activity
1K followers
-
Yamini Kagal reposted thisYamini Kagal reposted thisNVIDIA just made synthetic data free. The problem: → You need 100K training samples → Buying them costs $50K+ → Labeling them takes 6 months → Your model ships never The fix: → Generate training data from scratch → Augment your seed data at scale → Any model endpoint → Apache 2.0. Run it locally. What you can generate: → Instruction-tuning datasets → Domain-specific Q&A pairs → Evaluation benchmarks → Privacy-safe alternatives to production data 📦 pip install data-designer Not another wrapper. Not another API. Your data. Your GPU. Your rules. 💾 Save this for your next "we don't have enough data" blocker ♻️ Repost to save someone $50K and 6 months of labeling
-
Yamini Kagal reposted thisYamini Kagal reposted this🚀 We just open-sourced the NeMo Data Designer Library! Synthetic data has become one of our most powerful techniques at NVIDIA for generating Nemotron pre-training and post-training datasets at scale- and today, we’re making that same capability available to everyone. The NeMo Data Designer Library is now fully open source and pip-installable. It’s a flexible, scalable pipeline for high-quality synthetic data generation (SDG) built to remove data bottlenecks and accelerate AI customization and evaluation. 📦 [uv] pip install data-designer ⭐ Repo link and full details in the comments 👇 Our goal: empower builders everywhere to create better models, faster. Can’t wait to see what you build with it!
-
Yamini Kagal reposted thisYamini Kagal reposted thisCheck out how Adobe and CrowdStrike are accelerating their AI adoption using synthetic data with NeMo Data Designer! Adobe: https://lnkd.in/dW9Cmpv9 CrowdStrike: https://lnkd.in/dim8jc4E You can get started today (all you need is an API Key): https://lnkd.in/diaxXyvt
-
Yamini Kagal reposted thisYamini Kagal reposted this👉 Heading to AWS re:Invent 2024 this year? ➡️ Stop by our booth to discover how Gretel's synthetic data platform empowers developers to safely create realistic datasets from scratch or transform sensitive data into secure synthetic versions. Book a meeting in advance: https://lnkd.in/eKyctdwE 👋 See you there! #AWS #reInvent #AI #syntheticdata #dataprivacy
-
Yamini Kagal shared thisCheck out Data Designer – a powerful new feature within Gretel Navigator! If creating high-quality datasets to fine-tune models has been a challenge, Data Designer is here to help. Designed as a flexible tool for building realistic, diverse, and complex synthetic datasets, it includes pre-built blueprint templates to help get projects off the ground quickly. Our early preview launched today and includes blueprints for Text-to-Code (Python, SQL) datasets. Ready to see Data Designer in action? 🎉 Sign up now and help us shape the future of synthetic data!Yamini Kagal shared thisWe’re thrilled to announce the launch of Navigator Data Designer, now available in Early Preview! 🎉 This new feature allows you to generate realistic and diverse synthetic datasets to train your AI models. The first use case blueprint we’ve enabled with Data Designer is for building Text-to-Code (SQL, Python) datasets that mirror real-world coding scenarios and we have more use case blueprints coming soon - stay tuned! Here’s what you can expect from Gretel Navigator Data Designer: 💥 Boost AI Model Performance – Improve reasoning, generalization, and task handling with code training. ⚡ Accelerate AI/ML Development – Speed up model creation and evaluation by using high-quality synthetic datasets, reducing time-to-market for your AI initiatives. 💸 Reduce Costs – Generate reliable training and evaluation data at scale, reducing the high costs of sourcing and labeling datasets for AI model training. Want to dive deeper? Read our latest blog post (https://bit.ly/4etfhT6) to learn more, and don’t miss your chance to join the Early Preview. Sign up here: https://bit.ly/3AGKRit. We’re excited to see what you’ll build with Gretel Navigator Text-to-Code! #AI #syntheticdata #text-to-code #python #sql #ML
-
Yamini Kagal shared thisBoost healthcare AI with private, precise fine-tuning on sensitive text—powered by Gretel and AWS BedrockFine-tuning Models for Healthcare via Differentially-Private Synthetic TextFine-tuning Models for Healthcare via Differentially-Private Synthetic Text
-
Yamini Kagal shared thisCome work at Gretel! We’re looking for a Sr.PM to lead our partner integrations strategy. If you’re interested in driving impactful partnerships and being a core contributor to the growth and adoption of our generative AI technology, and you have a proven track record of thinking strategically while delivering tactically, we’d love to talk to you.
-
Yamini Kagal shared thisA milestone in our efforts to make synthetic data quality measurable and transparent: the Synthetic Data Utility Report. Excited for our customers to try it and to hear their feedback. Available in the Gretel Web Console now at https://console.gretel.ai/! Kudos to Nicole Pang for leading this release.Yamini Kagal shared thisWe're excited to announce the release of Gretel’s synthetic data utility report and machine learning quality score (MQS), which enables you to quickly understand how your synthetic data compares to real-world data when training ML models. 🥳 🎉 https://grtl.ai/MQSCompare Synthetic and Real Data on ML Models with the new Gretel Synthetic Data Utility ReportCompare Synthetic and Real Data on ML Models with the new Gretel Synthetic Data Utility Report
-
Yamini Kagal reacted on thisYamini Kagal reacted on thisNVIDIA Nemotron 3 Nano Omni brings text, image, video, and audio reasoning into a single open model, a big step toward making multimodal agent workflows much simpler and efficient. And it topped 6 leaderboards for complex document intelligence, and video and audio understanding. https://lnkd.in/gCx5y5vT Try it here: https://lnkd.in/gVA2Se9F Padmavathy S Bryan Catanzaro Jonathan Cohen Andrew Tao Chintan Patel Erik Pounds Karan Sapranemotron-3-nano-omni-30b-a3b-reasoning Model by NVIDIA | NVIDIA NIMnemotron-3-nano-omni-30b-a3b-reasoning Model by NVIDIA | NVIDIA NIM
-
Yamini Kagal liked thisYamini Kagal liked thisI really enjoyed the fireside chat between CEOs Jensen Huang and Shantanu Narayan at the Adobe Summit earlier this week. They share a vision for how technology supports creativity, and they also share a deep friendship that goes back nearly two decades. While listening to them on stage, everyone in the audience could feel the warmth and mutual respect. I was struck by the importance of personal relationships and humor amidst the warp speed of AI innovation. Seeing these visionary leaders laugh together as they signed the Sharpie booth was a highlight. Highly recommend spending 20 minutes with the replay if you’re interested in the intersection of AI and creativity. https://lnkd.in/gfmkE2ux #AdobeSummit #GenerativeAI #Leadership #NVIDIA #Adobe
-
Yamini Kagal reacted on thisCongratulations to my amazing wife on her presentation at the World Congress of Anesthesiologists!Yamini Kagal reacted on thisHad an incredible time attending and presenting at the 19th World Congress of Anesthesiologists in the beautiful city of Marrakech. Very grateful for the opportunity to share my doctoral work with international colleagues, and learn from our field’s experts. #WCA2026
-
Yamini Kagal liked thisYamini Kagal liked thisHow do you run autonomous agents safely? Our team has been thinking about this problem for a while at NVIDIA, and we just open sourced #OpenShell, the runtime environment for autonomous AI agents. It gives you sandboxed execution, policy-enforced access control, and private inference routing so you can run agents with real tools and real access, safely. OpenShell is in alpha and moving fast, but we want to get it into developers’ hands and see what you can build with it. Highlights: - 🙌 Open: Apache 2.0 license - 🔒 Safe by Default: Sandboxes start with minimal outbound access - 🧭 Explainable Control: Per-binary, per-endpoint, per-method policy with audit logs - 🤖 Agent Agnostic: Run Claude Code, Codex, OpenClaw, or your own agent with no code changes - 🧠 Private Inference: Route model traffic with privacy and cost controls - 💻 Developer Friendly: CLI, terminal UI, live policy updates, and remote sandbox support Get started here - 🤖 Blog: https://lnkd.in/g7VKi5Du Huge shoutout to our amazing team at NVIDIA. Special thanks to John Myers, Alexander Watson, Drew Newberry, Piotr Mlocek, Johnny Greco and Kirit Thadaka for diving deep into the weeds to help make this happen Feel free to reach out if you have questions or feedback. Happy building 🚀
-
Yamini Kagal reacted on thisYamini Kagal reacted on thisThis may come as a surprise to some, but NVIDIA is a massive open-source contributor. Not just in terms of models, frameworks, and tools, but #DATA as well! 🚀 Check out our latest blog post by Jane Polak Scowcroft, Will Jennings, Rebecca Kao, Annie Prasanna Surla, Leanna Chraghchian, and yours truly. We dive into our approach to open data and the "extreme co-design" principle—how we design all components together to eliminate bottlenecks at scale. We also share a few recent examples: 💚 Nemotron Pre- & Post-Training data stack 🤖 Physical AI – multimodal robotics and AV data 🌍 Nemotron Personas – population-scale synthetic datasets powering Sovereign AI globally. 🧬 La Proteina – a fully synthetic, atomistic protein dataset for biological modeling and drug discovery. ⚡ SPEED-Bench – A standardized benchmark for evaluating speculative decoding performance. 📄 NVDocs-v1 – synthetic QA triplets designed to train and evaluate RAG systems. 🧗 Nemotron-ClimbMix – A 400B-token pre-training dataset recently adapted as the default in nanochat by Andrej Karpathy. GPT-2 level model can now be trained in just 2 hours (!) on a single 8XH100 node, courtesy of this high-quality data. Links in comments below 👇
-
Yamini Kagal liked thisYamini Kagal liked thisGoing to NVIDIA #GTC next week and want to learn how synthetic data can help power specialized agents? The 🎨 NeMo Data Designer team will be leading a workshop on building high-quality, production-grade synthetic data pipelines for agent systems. In this hands-on workshop, we’ll walk through a production-grade synthetic data pipeline, from knowledge-grounded seeds to validated multi-turn agent trajectories with live tool use. 🔧 The workshop is designed for teams building agent systems that need to perform reliably across real-world use cases. 🗓️ Thursday, March 19 | 10:00 AM PDT | San Jose 🔗 https://lnkd.in/gvy-rcXR #NVIDIAGTC #AI #SyntheticData #AgenticAI #NeMoDataDesigner
-
Yamini Kagal liked thisYamini Kagal liked this🦞 Developers can run OpenClaw on NVIDIA DGX Spark, bringing powerful agentic workflows directly onto NVIDIA Grace Blackwell systems. The step‑by‑step playbook is now available: https://lnkd.in/e-vtsYwG
-
Yamini Kagal reacted on thisYamini Kagal reacted on thisI've been in a lot of negotiations. I don't remember losing many of them. I remember getting destroyed by Hannah Gordon. Ten-ish years ago, she was representing the San Francisco 49ers. I was with a delegation from Intel. We walked in confident. We walked out humbled. She was the youngest person in the room for sure. Probably the only woman. And she out-lawyered, out-strategized, and out-negotiated every one of us… while building a partnership that worked for everyone. We still got a deal. We just didn't get the deal we thought we were getting. Turns out that was just another Tuesday for her! I walked out thinking one thing: I'd like to work with her one day. Or work for her. That was ten years ago. Finally, I get to make good on that thought. Please join me in welcoming Hannah Gordon to the NVIDIA Inception team. Hannah brings 20+ years operating at the highest levels of sports, media, law, and consumer brands, including a decade with the 49ers where she helped lead billion-dollar negotiations, a $2B stadium project, and initiatives that set the standard for what great leadership looks like. She's also the author of SZN of CHANGE, her playbook for staying free, bold, and confident through every season. I'm inspired by folks who have more degrees than me, significantly smarter than me, usually much younger than me, and of course more hair than me. Every season, she's crushed it. This one will be no different. Welcome, Hannah. LFG!!!! NVIDIA NVIDIA for Startups Jensen Huang Hannah Gordon Will Koffel Jen Hoskins Skye Hart Sydney Sykes Tobias Halloran Ozzy Johnson Rebecca Nevin Jaclyn Jones Karissa Pagliero Alison Wagonfeld Mylene Mangalindan Laura Fay Kristin Major Madison Huang Serge Lemonde James L. Les Karpas Nate Barnett Mack Barton
-
Yamini Kagal reacted on thisCannot wait for next week to present with Meghana about DataRobot’s role in NVIDIA’s AI Factory deployment, the lessons learned and how we turned them into an enterprise ready Agent Workforce Platform for all!Yamini Kagal reacted on thisNEXT WEEK: If you’ll be at #NVIDIAGTC, come say hi! At our breakout session, we’ll be joined by NVIDIA AI to learn how NVIDIA Enterprise AI Factory powers a versatile, full-stack ecosystem, where specialist platforms provide the backbone for running inference and agents in production. We’ll share patterns for deploying AI at scale—when to build vs. buy, how to work with software partners, and how to keep efforts tied to real ROI instead of science projects. Add it to your agenda: https://bit.ly/4b6FuIK
Recommendations received
4 people have recommended Yamini
Join now to viewOther similar profiles
Explore more posts
-
Sean Hardy
Most teams don’t struggle with AI agents because the technology is immature. They struggle because prompting is not treated with design discipline. An agent prompt is not a single instruction. It’s a system of assumptions, constraints, and decision rules. When those aren’t explicit, reliability breaks down fast. That’s why two teams can use the same models and get completely different outcomes. This guide breaks down how to design agent prompts the way operators design systems. Intentionally, repeatably, and with scale in mind. Full breakdown here: https://lnkd.in/gzhuQgRW #Agents #Prompting #AI
13 Comments -
Boaz Ashkenazy
Commerce is being quietly reinvented by agents that actually know the customer. In a recent conversation with Gordon Evans, CMO of Commerce Cloud at Salesforce, we talked about what happens when AI agents are powered by real data—order history, inventory, location, preferences—not just prompts. His framing stuck with me: the future of shopping comes down to two things—conversation and context. And that applies inside the office just as much as it does in the cart. Full conversation on the Shift AI Podcast.
2 Comments -
Sid Pardeshi
The design-to-code handoff is one of the most expensive gaps in enterprise software development. Weeks of engineering time spent interpreting Figma files, reconciling spacing math, and rebuilding what designers already specified. On March 30, join Aditya Karra, Blitzy's Founding Designer, and Ben Bergerson, Senior AI Solutions Consultant, for a walk thru of Blitzy native Figma-to-production capability. No prototypes. No Refine PR. Production-ready code. - Every design token read before a single line of code is written - Responsive output across five viewports from one desktop frame - A new feature integrated into an existing codebase, brand-aligned, checked directly into the repo Designers and product teams can now move at the speed of design. Come see what that looks like in practice. Register with the link in the comments. March 30 | 12pm ET | Free with live Q&A #Blitzy #Webinar #Figma #Platform40 #FrontendDevelopment #EnterpriseAI #ProductVelocity
1 Comment -
Sean Falconer
In the consumer world, we want a single interface that can write a poem, plan a vacation, and debug code. In that open world, scale is the only strategy that works. But enterprise workflows don’t live in an open world. Most B2B problems like parsing invoices, routing tickets, classifying clauses operate in closed systems. They have well-defined inputs, explicit outputs, and hard failure modes. In my latest article, I argue that the future of enterprise AI isn't always about getting bigger, it's often about model specialization. The most effective architectures I’m seeing are cascading. They use SLMs for routine volume, and escalate to LLMs only when deep reasoning is required. Read the full breakdown below. https://lnkd.in/gkyJbFWb
6 Comments -
Ben Skal
Industrial organizations have typically applied AI to their most critical assets and processes, the top 10%. In order to start expanding AI insights and recommendations to the remaining 90% of the facility, we need to lower the cost and effort to deploy and maintain these insights. The work we're doing with NVIDIA and Celanese to address this opportunity by grounding models on trusted, contextualized data is another exciting step in that direction.
1 Comment
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content