--- title: "What do a kid's party and building reliable systems have in common? Controlled chaos" newsletter: "MLOps Community" date: 2024-06-06 source: https://aaif.live/newsletters/mlopscommunity/2024-06-06-what-do-a-kid-s-party-and-building-reliable-systems-have-in --- # What do a kid's party and building reliable systems have in common? Controlled chaos *Plus, a dynamic duo talk about a dynamic duo, the new synthwave, drugs, and hidden gems.* *MLOps Community — Agentic AI Foundation, 2024-06-06* ## Build Reliable Systems with Chaos Engineering // Benjamin Wilms // MLOps Podcast #237 If you think chaos engineering is me setting up a pipeline, you need to listen to this episode. Benjamin told me how chaos engineering has developed from the chaos monkey tool to a more sophisticated, experiment-driven approach, and helps improve systems by intentionally injecting faults in a controlled, pre-production environment to assess risk and enhance reliability. He also highlighted the need for a cultural shift to prioritize reliability over speed, and stressed the importance of human insight in designing experiments and integrating into the CI/CD pipeline and observability tools to differentiate between induced conditions and organic system behaviors. Now, when I get things wrong, I can just say I was chaos engineering. Video [https://home.mlops.community/home/videos/build-reliable-systems-with-chaos-engineering] || Spotify [https://open.spotify.com/episode/0979z1OqYQAMmLceQNNN77?si=HxUwB6B1SnqQcKEFoAlGCQ] || Apple [https://podcasts.apple.com/us/podcast/build-reliable-systems-with-chaos-engineering-benjamin/id1505372978?i=1000657436509] [https://podcasts.apple.com/us/podcast/build-reliable-systems-with-chaos-engineering-benjamin/id1505372978?i=1000657436509](https://podcasts.apple.com/us/podcast/build-reliable-systems-with-chaos-engineering-benjamin/id1505372978?i=1000657436509) [https://home.mlops.community/home/videos/build-reliable-systems-with-chaos-engineering](https://home.mlops.community/home/videos/build-reliable-systems-with-chaos-engineering) [https://open.spotify.com/episode/0979z1OqYQAMmLceQNNN77?si=HxUwB6B1SnqQcKEFoAlGCQ](https://open.spotify.com/episode/0979z1OqYQAMmLceQNNN77?si=HxUwB6B1SnqQcKEFoAlGCQ) ## Infer Summer '24 ☀️ A free virtual conference from Qwak https://www.qwak.com/infer/summer-2024 We’re excited to invite you to Infer Summer, Qwak’s free virtual conference on June 26, 11:00AM EDT. This live event offers a unique opportunity to interact with experts, ask questions, and gain practical insights from real-world implementations. Join us to learn practical takeaways, including: * How to create a robust recommender system with a focus on architecture and user engagement strategies using advanced machine learning models. * The construction techniques used to develop the world's first AI-based sports narrator, and how to implement solutions that provide predictive outcomes with reasoning. * How to address risks of inaccurate and harmful content in AI systems, implement "Schematic Questioning", and navigate regulatory challenges to deliver secure, customer-centric AI solutions. Hear directly from top industry professionals, including speakers from Disney Streaming, Lightricks, LSports, and Lili Banking, who will provide actionable insights. Register here for free [https://www.qwak.com/infer/summer-2024?utm_source=mlops_community&utm_medium=newsletter&utm_campaign=infersummer24] - We look forward to seeing you at Infer Summer! 😎 ## AWS Tranium and Inferentia // Kamran Khan and Matthew McClean // MLOps Podcast #238 We don’t have Batman and Robin, but this episode is all about dynamic duos - Matt and Kamran on to talk about AWS's Inferentia and Tranium. We get into their development, their impact on AI workloads, and how AWS designed these AI accelerators to offer better performance and lower costs than traditional GPUs. Inferentia speeds up inference, while Tranium is built for training large models and we talk about their integration with PyTorch and TensorFlow using the Neuron SDK. They also share some real-world examples like partnerships with Anthropic to show how these accelerators reduce costs and improve efficiency, with deployment options in Sagemaker, Bedrock, and Kubernetes. Clicking below and giving a like – another powerful duo! Video [https://home.mlops.community/home/videos/aws-tranium-and-inferentia] || Spotify [https://open.spotify.com/episode/2Y1wLwi6fzH4jmu6NuLMYO?si=56TqgZWFTgeaOd9EU_7tBQ]|| Apple [https://podcasts.apple.com/us/podcast/aws-tranium-and-inferentia-kamran-khan-and/id1505372978?i=1000657825953] [https://podcasts.apple.com/us/podcast/aws-tranium-and-inferentia-kamran-khan-and/id1505372978?i=1000657825953](https://podcasts.apple.com/us/podcast/aws-tranium-and-inferentia-kamran-khan-and/id1505372978?i=1000657825953) [https://home.mlops.community/home/videos/aws-tranium-and-inferentia](https://home.mlops.community/home/videos/aws-tranium-and-inferentia) [https://open.spotify.com/episode/2Y1wLwi6fzH4jmu6NuLMYO?si=56TqgZWFTgeaOd9EU_7tBQ](https://open.spotify.com/episode/2Y1wLwi6fzH4jmu6NuLMYO?si=56TqgZWFTgeaOd9EU_7tBQ) ## Upcoming Mini-Summit with Zilliz https://home.mlops.community/home/events/practical-approaches-to-unstructured-data-using-vector-databases-and-rag-2bucmb4j1s?agenda_day=664cd9264120d5c7f5f4714e&agenda_track=664cd9274120d5c7f5f47160&agenda_stage=664cd9264120d5c7f5f47154&agenda_filter_view=stage&agenda_view=list Vexed by vectors? Raging with RAG? Mad with multimodal? Join us 12 June for our live mini-summit featuring experts from Zilliz and Jina AI, where we’ll explore practical approaches to using vector databases, RAG, and multimodal embeddings. Learn how RAG systems are revolutionizing natural language processing by integrating information retrieval, discover strategies for maintaining data freshness and efficiency, and how multimodal embeddings can transform data interaction. Don't miss the chance to ask questions and interact with leading voices in the field. Register for free here [https://home.mlops.community/home/events/practical-approaches-to-unstructured-data-using-vector-databases-and-rag-2bucmb4j1s?agenda_day=664cd9264120d5c7f5f4714e&agenda_track=664cd9274120d5c7f5f47160&agenda_stage=664cd9264120d5c7f5f47154&agenda_filter_view=stage&agenda_view=list] ## Job of the Week [https://www.hoplabs.com/ml-software-engineer](https://www.hoplabs.com/ml-software-engineer) ## Synthetic Data for Computer Vision // Rich Riley // MLOps Community IRL Meetup #80 Bristol ## MLOps Community IRL Meetup Synthwave v2.0 is coming, but it's not music, it's data. Rich discusses the benefits of using synthetic data in machine learning, particularly for computer vision tasks. He covers how synthetic data is created and addresses data labeling complexities and methods to generate synthetic data, like 3D modeling and generative AI. He share its advantages over manually annotated data, highlighting how it removes bottlenecks in machine learning, and offers insights into future technologies, such as Nvidia's Isaac Sim and generative models like Dali and Stable Diffusion. He also gives a practical walkthrough on generating synthetic datasets using Blender for drone detection and discusses the challenges of bridging the gap between synthetic and real data. Dust off your shoulder pads and get ready for the new wave. Watch it here [https://home.mlops.community/home/videos/synthetic-data-for-computer-vision] [Watch it here](https://home.mlops.community/home/videos/synthetic-data-for-computer-vision) ## The MLOps Behind Recursion’s Foundation Model Phenom-1 I know we’ve got our merch with the drug references, we’ve now branched out to blogs too. This is all legit though, as it’s a look at Recursion’s phenomics approach to drug discovery. It uses a deep learning model trained on 95 million microscopy images to analyze images of genetically or chemically perturbed cells and create ‘maps of biology’. They share details of the foundation model, including the use of ViT architecture with a masked autoencoder approach for self-supervised pre-training, data preparation using the Zarr format, and the compute infrastructure employed. Plus they cover the inference pipeline, which involved optimizations such as TensorRT conversion and collaboration with NVIDIA and GCP to increase throughput. I wonder if they’ll discover some new merch potential? With thanks to Vasudev Sharma, Ayla Khan, Jess Leung and Kristen Morse for their contribution. Creating a PDF Query Assistant with Upstage AI Solar and LangChain Integration [https://home.mlops.community/home/blogs/creating-a-pdf-query-assistant-with-upstage-ai-solar-and-langchain-integration] Do you find PDF often stands for Pretty Damn Frustrating? Check out this blog that explains how to build a PDF query assistant using LangChain and Upstage AI Solar. It covers integrating LangChain for text extraction, using OpenAI embeddings for semantic analysis, and implementing a vector database with Chroma for efficient querying. Additionally, it details the use of text splitters for managing document chunks and the RetrievalQA system for interactive querying, enabling efficient data extraction and analysis from PDFs. Once you’ve read it, it’ll be Perfectly Deciphered Files all the time. With thanks to Sonam Gupta for their contribution. [The MLOps Behind Recursion’s Foundation Model Phenom-1](https://home.mlops.community/home/blogs/the-mlops-behind-recursions-foundation-model-phenom-1) ## Hidden Gems ## IRL Meetups San Francisco [https://www.meetup.com/mlops-community/events/301277494/?utm_medium=referral&utm_campaign=share-btn_savedevents_share_modal&utm_source=link] - June 10 - 🙌thanks to Tecton Frankfurt [https://www.meetup.com/frankfurt-mlops-community/events/300877018/?utm_medium=referral&utm_campaign=share-btn_savedevents_share_modal&utm_source=link] - June 12 Denver [https://www.meetup.com/denver-mlops-community/events/300432149/?utm_medium=referral&utm_campaign=share-btn_savedevents_share_modal&utm_source=link] - June 12 - 🍻cheers to Weights & Biases Stockholm [https://www.meetup.com/stockholm-mlops-community/events/301361666/?utm_medium=referral&utm_campaign=share-btn_savedevents_share_modal&utm_source=link] - June 18 - 📣shoutout to HPE London [https://www.meetup.com/reproducibility-and-productivity-in-data-science/events/301219107/?utm_medium=referral&utm_campaign=share-btn_savedevents_share_modal&utm_source=link] - June 19 San Francisco [https://www.meetup.com/mlops-community/events/300375509/?utm_medium=referral&utm_campaign=share-btn_savedevents_share_modal&utm_source=link] - June 25 - 🙏with Kolena [San Francisco](https://www.meetup.com/mlops-community/events/301277494/?utm_medium=referral&utm_campaign=share-btn_savedevents_share_modal&utm_source=link) Thanks for reading. See you in Slack [https://go.mlops.community/slack], YouTube [https://www.youtube.com/channel/UCG6qpjVnBTTT8wLGBygANOQ?view_as=subscriber], and podcast [https://home.mlops.community/public/content/] land. Oh yeah, and we are also on X [https://twitter.com/mlopscommunity]. The MLOps Community newsletter is edited by Jessica Rudd [https://www.linkedin.com/in/jmrudd/]. --- Source: https://aaif.live/newsletters/mlopscommunity/2024-06-06-what-do-a-kid-s-party-and-building-reliable-systems-have-in