Elena Bonfiglioli’s Post

The Question Every Organization Must Ask: Is Your Data Ready for AI? Many companies begin their AI journey with a goal — to explore how artificial intelligence can transform their business. Yet, along the way, they discover a crucial truth: their data isn’t ready for AI. The companies leading in AI today aren’t just investing in algorithms; they’re investing in data they’re investing in data readiness: ✅ Governance ✅ Interoperability ✅ Trust across the entire data estate Before you build your next AI model or product, pause and ask yourself: 🧠 Is my organization’s data truly AI-ready? #AIReadiness #DataStrategy #DigitalTransformation #AzureAI #AzureOpenAI #MicrosoftFabric #Databricks #AWS #SageMaker #GoogleCloud #VertexAI #EnterpriseAI #AnalyticsToAI #DataDriven #DataGovernance #CloudDataPlatform #GenAI #ArtificialIntelligence #MachineLearning #DataEngineering #MLOps #AIFirst

• 𝐆𝐏𝐓-5-𝐏𝐫𝐨 delivers advanced reasoning and analytics through a tournament-style architecture—ideal for complex analytics, code generation, and intelligent decision-making. • 𝐆𝐏𝐓-5-𝐂𝐡𝐚𝐭 now includes enhanced safety guardrails to better support users during sensitive conversations, reinforcing our commitment to responsible AI. • 𝐆𝐏𝐓-𝐈𝐦𝐚𝐠𝐞-1-𝐦𝐢𝐧𝐢, 𝐆𝐏𝐓-𝐫𝐞𝐚𝐥𝐭𝐢𝐦𝐞-𝐦𝐢𝐧𝐢, 𝐚𝐧𝐝 𝐆𝐏𝐓-𝐚𝐮𝐝𝐢𝐨-𝐦𝐢𝐧𝐢 are now available, enabling real-time image, voice, and audio generation with minimal infrastructure. These models are optimized for speed, affordability, and seamless integration into your existing workflows. • 𝐒𝐨𝐫𝐚 2 is coming soon to Azure AI Foundry, bringing advanced video and audio generation in a single API. Think physics-driven animation, synchronized dialogue, and dynamic media creation. Layer in the 𝐌𝐢𝐜𝐫𝐨𝐬𝐨𝐟𝐭 𝐀𝐠𝐞𝐧𝐭 𝐅𝐫𝐚𝐦𝐞𝐰𝐨𝐫𝐤 and 𝐧𝐞𝐰 𝐦𝐮𝐥𝐭𝐢-𝐚𝐠𝐞𝐧𝐭 𝐨𝐫𝐜𝐡𝐞𝐬𝐭𝐫𝐚𝐭𝐢𝐨𝐧 𝐜𝐚𝐩𝐚𝐛𝐢𝐥𝐢𝐭𝐢𝐞𝐬, and you’ve got a platform built to solve real business challenges, whether that’s automating customer service or delivering content at global scale. AI is no longer just something to experiment with. It’s now about putting it into action—deploying real solutions that reduce latency, cut costs, and speed up time to value.  Learn more and get started: http://aka.ms/AOAIOCT

#OpenAI and #Databricks partnering on models is not just another integration—it’s a step toward making enterprise AI more usable and trustworthy. #Databricks brings governed, high-quality data and compute at scale. #OpenAI brings cutting-edge models. Together, they enable enterprises to move from experimentation to production—where models are fine-tuned, context-aware, and grounded in real business data. This is important because the gap between AI research and business impact has always been operationalization. With Databricks + OpenAI, enterprises get a platform where governance, cost efficiency, and accuracy are built in. In simple terms: enterprises can now use AI instead of just talking about it.

AI agents that perform better than proprietary models, at 90× lower cost! That’s what Databricks just showed is possible with their latest work on enterprise AI agents. Instead of spending quite a time fine-tuning models or burning through inference dollars, they use 'automated prompt optimization' to systematically search for the best prompts, guided by evaluation signals. The outcome: open-source models, tuned this way, beat some of the most advanced proprietary models, while making enterprise-scale deployment affordable. For anyone building LLM agents, this means: 1. Higher accuracy and reliability 2. Dramatically lower inference spend 3. Faster iteration without retraining models Under the hood, this uses structured search (their GEPA optimizer) to explore large number of prompt variants, then selects the ones that maximize performance. The cost gets distributed across all future requests. This could redefine how we approach prompt engineering and make high-quality AI more accessible for production systems. Read more here: https://lnkd.in/g8xut4N4

🚀 Just explored Microsoft's latest post on Azure AI Foundry: "How Microsoft Evaluates LLMs: A Practical, End-to-End Playbook" A goldmine for anyone building modular, scalable, and responsible AI systems. 🔍 Key takeaways: Evaluation as a pipeline: Covers everything from prompt design to telemetry, not just model metrics. Multi-layered testing: Combines automated evals, human-in-the-loop feedback, and real-world scenario simulation. Tooling-first mindset: Emphasizes reproducibility, traceability, and integration with Azure ML workflows. Open-source alignment: Built to complement OSS agents, RAG systems, and orchestration frameworks. 💡 For those designing agentic systems or teaching best practices in LLM evaluation—this is a must-read. 📎 Read the full playbook https://lnkd.in/gg9AiN5W #AzureAIFoundry #LLMEvaluation #ResponsibleAI #CHAI #BenchmarkingPlaybook #AgenticSystems #ReproducibleAI #tac

Just caught wind of this; Databricks and OpenAI are teaming up in what they’re calling a multi-year, $100M #partnership to bring frontier AI into the enterprise, natively, on Databricks’ own #platform. What’s exciting to me is how this moves the goalpost: • OpenAI models will be integrated with Databricks, allowing enterprises to leverage AI without moving their data. • Databricks’ Agent Bricks will enable organisations to build AI agents with built-in governance, evaluation, and iteration. • The partnership emphasises security, observability, and ethical standards for deploying AI at scale. For me, this feels like a turning point. The “AI in the wild” era is maturing into “AI as part of the backbone.” When AI agents can be built securely on enterprise data, without friction, that’s where real change happens. Curious to see: • Will this setup deliver the performance we expect? • How much control will businesses retain over tweaks, tuning, and oversight? • And, how soon will enterprises adopt this as standard rather than experimental? If you’re working in data, AI, or building products, this is one to watch. #AI #EnterpriseAI #DataIntelligence #OpenAI #Databricks #Innovation #AITransformation

The Databricks AI research team just shared some exciting results: automated prompt optimization can reach the same quality levels as supervised fine-tuning, while keeping serving costs lower. Efficiency and performance don’t have to be a trade-off anymore.

GPT-5 Model Family Now Powers Azure AI Foundry Agent Service.   The GPT-5 family of models is now available in Azure AI Foundry Agent Service, bringing Azure OpenAI’s most advanced reasoning, multimodal, and coding intelligence into a platform purpose-built for enterprise-scale agent development.   The new generation of models With this release, developers can choose from the full GPT-5 lineup to balance speed, depth, and cost for their agent development needs: GPT-5 — the flagship, with a 272k-token context window, designed for deep analysis, complex automation, and high-trust scenarios such as analytics and compliance. GPT-5-mini — fast and efficient, ideal for real-time interactions and reliable tool use. GPT-5-nano — ultra-low latency and cost-optimized for high-volume requests and lightweight orchestration. GPT-5-chat — a multimodal specialist with a 128k-token context window, enabling natural conversation and contextual... #techcommunity #azure #microsoft https://lnkd.in/g4j86y-Q

⚡ AWS just announced a major update to Bedrock: now you can fine-tune foundation models with your own data. As someone who’s been experimenting with AWS’s AI services, this is a game-changer. Fine-tuning lets you adapt powerful models to your specific use case—whether it’s improving chatbots, automating content generation, or enhancing recommendation systems. I’ve been playing with Bedrock for text generation, and this update makes it even more versatile. The ability to train models on proprietary datasets means we can finally bridge the gap between generic AI and domain-specific intelligence. What’s your take on fine-tuning vs. prompt engineering? Do you see this as a turning point for enterprise AI adoption? 👇 #AI #AWS #MachineLearning #CloudComputing #DataScience

In my latest blog post, I explore how Snowflake Cortex helps close the gap between AI pilots and achieving AI driven outcomes by enabling teams to run AI directly on Snowflake data, use both built-in and custom LLM capabilities without adding new infrastructure, and power use cases like document intelligence and knowledge-based chat. https://lnkd.in/g-jkBtaH