Nicolas Gaudilliere’s Post

#ai #redhat Red Hat went all-in on generative AI at its annual summit, offering a wide range of tools for operational and development teams to help them build and deploy generative AI systems. That includes tools for creating and managing a model garden, training and fine-tuning models, building applications, and deploying generative AI at scale in a hybrid architecture.

Generative AI is entering its agents era. The agentic architectures and early examples we see today represent only the beginning of a broader transformation that promises to redefine the human-machine dynamic, with implications for both enterprise applications and infrastructure. #ai #aiagents #agenticAI

💡 Unlock the Power of #GenerativeAI for #MLOps 💡 The future of AI is here with Generative AI Operations (#GenAIops), enhancing MLOps workflows and driving innovation across industries. 🚀 🔍 What’s New? - Seamless integration of Generative AI into MLOps pipelines. - Enhanced model lifecycle management for AI-driven solutions. - Automation of tasks like model training, deployment, and monitoring to scale AI efforts. Generative AI isn’t just transforming the way we build AI models-it’s reshaping how we operationalize them for maximum impact. #AI #GenerativeAI #MLOps #AzureAI #AIInnovation #Automation #Expert #microsoft Sword Luxembourg Experience Microsoft

Want to learn more about the real power of 💬Generative AI + 🕸️Knowledge Graph on Amazon Neptune? Please take a look on my co-authored solution guidance and sample code for the manufacturing digital thread demo on AWS✍️. https://lnkd.in/g5TQG68r 🌟Special thanks to Raja GT and Vedanth Srinivasan for contirbuting to the guidance. #aws #GenerativeAI #AmazonBedrock #AmazonNeptune #GraphDatabase #SolutionGuidance #SampleCode #manufacturing

To all those planning to build a GenAI application: context is the cornerstone of relevant, meaningful interactions. However, the way AI models handle context varies significantly, depending on the provider's architecture. Context refers to an AI’s ability to remember prior exchanges. For instance, if you’re using a virtual assistant, you wouldn’t want to repeat details every time you ask a follow-up question. Managing this context within the application or burdening the user to repeat themselves is inefficient. There are two distinct approaches to managing context in AI systems: 1. Rich Context Management (Threading): Some models, like those from OpenAI and Azure, offer threading capabilities. These models retain memory throughout the conversation, enabling smooth, natural exchanges. This eliminates the need to resend context with every request, making them perfect for chatbots and other systems where maintaining flow is crucial. Additionally, it optimizes costs, as you avoid repeated charges for sending the same information. 2. Low Context Management (Re-send Context): Other models, like those from Anthropic and AWS Bedrock, require the full context to be sent with each request to maintain relevance. Though this can lead to additional token charges, it works well for expert systems, where each interaction stands alone, and precise, one-off analysis is the goal. Long-term memory isn’t always necessary, especially for models performing independent tasks. Choosing the right model depends entirely on your application’s needs. If your system relies on seamless, continuous interaction—such as in customer support—models with rich context management are the right choice. Conversely, for scenarios requiring detailed, standalone responses, where each request is independent, models requiring re-sent context might serve you better. By understanding these distinctions, you’ll be able to select the most suitable GenAI model and provider for your application. #AI #ContextManagement #ExpertSystems #TechnologyLeadership #ConversationAI #GenAI

Gemini 1.5 release could be a breakthrough in enterprise use cases. Will it blow RAG out of the arena? The new Gemini 1.5 got buried a little because of the more flashy Open AI Sora announcement, but it’s worth digging into if you are working in Enterprise applications in generative AI. The headline is that it uses a million token context window. This in itself should garner a healthy dose of skepticism as larger context windows often suffer with the “needle in the haystack “ problem. In other words the models can ingest huge documents but really only can handle context at the beginning and end of the documents. The paper mentions that this could reduce or even completely remove the need to use RAG. We have heard this claim before with the release of Claude and many others. This blog and attached paper claims to be using a new version of the Mixture of Experts (MoE) architecture which massively improves the haystack problem among other things. This of course this the type of architecture used by Mistral AI but recent papers have shown a leap forward in this type of architecture. There is also a lot of impressive multi modal features such as precision video queries. The blog and videos are worth some time if you find yourself constantly struggling with vector databases and RAG. I am skeptical but definitely keeping an eye on it. https://lnkd.in/e5jHnvYQ

Did you know that Generative AI can do much more than just summarization and Copilot type use cases? Check out this informative blog post that covers five real-world examples of how generative AI is being used today, complete with architecture examples. If you're looking to bring this cutting-edge technology to your agency / organisation, there's no better partner than Microsoft. Follow the link to learn more! Link: https://lnkd.in/gruii75X

New AWS Solution Guidance Released: Building a Digital Thread with Graph and Generative AI on AWS 🚀 We are excited to share the latest Solution Guidance which demonstrates how to create an intelligent manufacturing digital thread through a combination of knowledge graph (#AmazonNeptune) and Generative AI (#AmazonBedrock #AnthropicClaude) on AWS. The guidance is now available here: https://lnkd.in/g6fRjZbF This guidance is tailored for manufacturers keen on improving their processes through graph database and Generative AI for manufacturing digital thread. By integrating knowledge graph and generative AI, manufacturing organizations can enhance data integration, improve semantic understanding, and enable intelligent and context-aware applications leading to a more personalized user experience. 🔎 What's Inside: Solution Architecture: Explore the solution architecture outlined in this guidance, providing an approach to seamlessly integrate graph database and Generative AI into the manufacturing digital thread. Enhance Digital Thread relationships using Amazon Neptune graph database by creating a connected and streamlined digital thread experience for your operations. Enhance user experiences using Amazon Bedrock to understand complex digital thread graph data, analyze relationships, and provide valuable insights in natural language. Solution Implementation: Access the GitHub repository linked with this guidance for the code, featuring one-click deployment model that effortlessly creates the sample manufacturing digital thread demo on AWS. https://lnkd.in/gct-D7Kn Contributors: Special thanks to Vedanth Srinivasan and Shing Poon for contributing to the guidance. #DigitalThread #Manufacturing #AWSManufacturing #Industry40 #GraphDatabase #GenerativeAI #ConnectedData #ManufacturingInsights #AWSNeptune #AmazonNeptune #AmazonBedrock

**🌐 Implementing a Scalable AI Monitoring Pipeline for Real-Time Model Drift Detection** [Designed on CodeLlama - runs on GPT4o, Gemini Pro, Grok] **The Challenge:**   As our operations expanded across multiple regions, we faced the growing problem of model drift—where AI models gradually lose accuracy due to evolving data. Manual detection and retraining were insufficient at this scale, leading to inefficiencies and delayed responses. **The AI Solution:**   We designed an **automated monitoring pipeline** using **TensorFlow Data Validation (TFDV)** for data schema validation and drift detection, combined with **Apache Kafka** for real-time streaming. This pipeline tracks data changes in real-time, using statistical tests like **Kolmogorov-Smirnov** and **Chi-square** to detect drift. Alerts are triggered through **Grafana and Prometheus**, and the system initiates region-specific retraining when necessary, ensuring continued model performance across all environments. **The Results:**   This implementation has led to a **50% reduction in drift-related incidents** and a **30% improvement in model accuracy** across regions. The pipeline now autonomously detects drift, alerts teams, and retrains models as needed, significantly reducing manual intervention and optimizing overall system performance. **The Future of AI Operations:**   AI is becoming essential for maintaining accuracy and scalability in modern operations. Our experience underscores how automated, AI-driven pipelines can revolutionize efficiency and decision-making. As the technology evolves, businesses that adopt such systems will continue to innovate faster and lead their industries. https://lnkd.in/ggQyr2vX #AI #ModelDrift #MachineLearning #TensorFlow #ApacheKafka #RealTimeMonitoring #Automation #AIinOperations #DataValidation #ai #innovation #monitoring #devops #mlops #drift

Ready to explore how AI can boost your business? 📈 VIAcode’s free AI Proof of Value is the perfect starting point! Backed by Microsoft so there is no cost to you, VIAcode will develop and deliver an AI solution for your business. Don’t miss this chance to experience the real value AI can bring to your business. 🔗 Get started today: https://lnkd.in/eQXWeyZu #AI #Microsoft #AIPilot #BusinessSolutions #Azure #Free #BusinessGrowth #LinkedInLearning