--- title: "You know quality when you see it. But apart from the mirror, what does it look like?" newsletter: "MLOps Community" date: 2024-05-09 source: https://aaif.live/newsletters/mlopscommunity/2024-05-09-you-know-quality-when-you-see-it-but-apart-from-the-mirror-w --- # You know quality when you see it. But apart from the mirror, what does it look like? *Plus, transitioning to GenAI, don't get in to debt, getting a hit of MMDB, and hidden gems* *MLOps Community — Agentic AI Foundation, 2024-05-09* Following the collapse of ArcaVacua™ due to one bad review, we’re launching our new start-up. Ever wondered what problems all the latest AI products and gimmicks are meant to solve? Well wonder no more! USE302™ is our fully automated, AI driven, open-source API, designed to find the problems that didn’t exist until AI was designed to solve them. It harnesses the power of blockchain technology, microservices architecture, and quantum computing to deliver a scalable, disruptive, and innovative solution that transforms the way you leverage data from empty buzzword solutions to productionized profits. We’re welcoming investment now, with a guaranteed ROI* *Regret on investment ## What is AI Quality? // Mohamed Elgendy // MLOps Podcast #229 It's important to have standards. Take the quality of jokes in this newsletter for example—there are standards, they’re just very low. Of a much higher standard is this chat with Mo about the importance of AI quality, and the push for Gold Standards. It all started with a comment from a colleague, “We are providing testing to our customers, we're not providing quality”. This led to them thinking "What is quality?” - and from there it’s led to us setting up AIQCON [https://www.aiqualityconference.com/] together in June. We want folk to have fun and learn things on the day, but we also want to have tangible outcomes, like working groups to help move things forward. As Mo explains, the conference is just the start of a three-stage plan: 1. Domain Discovery: Pinpoint areas of AI that need clear definitions of AI quality standards. 2. Collaboration: AI builders, regulators, and experts working together to define risk and application guidelines. 3. Updates and Releases: Continual revisions of standards to ensure they remain relevant. Standard closing line: click below and give it a watch! Video [https://home.mlops.community/home/videos/what-is-ai-quality] || Spotify [https://open.spotify.com/episode/4QnLA4XUmTn9Lo8Bjq63YT?si=8WDUEKCAQRiZMI-q7MDybA] || Apple [https://podcasts.apple.com/us/podcast/what-is-ai-quality-mohamed-elgendy-228/id1505372978?i=1000654460381] [https://podcasts.apple.com/us/podcast/what-is-ai-quality-mohamed-elgendy-228/id1505372978?i=1000654460381](https://podcasts.apple.com/us/podcast/what-is-ai-quality-mohamed-elgendy-228/id1505372978?i=1000654460381) [https://home.mlops.community/home/videos/what-is-ai-quality](https://home.mlops.community/home/videos/what-is-ai-quality) [https://open.spotify.com/episode/4QnLA4XUmTn9Lo8Bjq63YT?si=8WDUEKCAQRiZMI-q7MDybA](https://open.spotify.com/episode/4QnLA4XUmTn9Lo8Bjq63YT?si=8WDUEKCAQRiZMI-q7MDybA) ## FEDML Nexus AI: Your Generative AI Platform at Scale // Salman Avestimehr // MLOps Podcast #230 A big part of these podcasts is that you learn something, then transition to generating amazing things. There's a certain symmetry to that in this episode as Salman discusses the development of FEDML. Initially focused on federated learning, they've broadened their scope to become a multi-cloud GenAI platform for managing model serving for scalability, cost, and observability. He also breaks down the lifecycle of an AI model, particularly addressing two challenges that AI developers and enterprises face in their GenAI pipeline: ownership and scalability. Learn more by generating a click to listen! Video [https://home.mlops.community/home/videos/fedml-nexus-ai-your-generative-ai-platform-at-scale] || Spotify [https://open.spotify.com/episode/5Rm7qg0Tot4MQ3R63jKg11?si=tfH3SY5uQ7qL4M6X1nsHEQ] || Apple [https://podcasts.apple.com/us/podcast/fedml-nexus-ai-your-generative-ai-platform-at-scale/id1505372978?i=1000654824082] [https://podcasts.apple.com/us/podcast/fedml-nexus-ai-your-generative-ai-platform-at-scale/id1505372978?i=1000654824082](https://podcasts.apple.com/us/podcast/fedml-nexus-ai-your-generative-ai-platform-at-scale/id1505372978?i=1000654824082) [https://home.mlops.community/home/videos/fedml-nexus-ai-your-generative-ai-platform-at-scale](https://home.mlops.community/home/videos/fedml-nexus-ai-your-generative-ai-platform-at-scale) [https://open.spotify.com/episode/5Rm7qg0Tot4MQ3R63jKg11?si=tfH3SY5uQ7qL4M6X1nsHEQ](https://open.spotify.com/episode/5Rm7qg0Tot4MQ3R63jKg11?si=tfH3SY5uQ7qL4M6X1nsHEQ) ## Job of the Week [https://voyant.bio/#roles](https://voyant.bio/#roles) ## AI Innovations: The Power of Feature Platforms // MLOps Mini Summit #6 ## MLOps Community Mini Summit Mario, Sonic, and Crash Bandicoot. 3 icons of platformers. This mini summit doesn’t have those, but it does have Mahesh, Jose, and Nikhil, 3 icons of feature platforms. They're here with a great set of talks about their struggles and triumphs in streamlining feature engineering and implementation. They share insights into overcoming infrastructural hurdles and improving model performance. We even get a look at Tecton's new platform, Rift, which helps data scientists with Python simplicity and enterprise-grade performance. I feel like with that intro, I can just about get away with the cliché of 'level up your game'! Watch it here [https://home.mlops.community/home/videos/ai-innovations-the-power-of-feature-platforms-mlops-mini-summit-6] ## Technical Debt in ML Systems // Francesca Carminati // IRL #76 Stockholm ## MLOps Community IRL Meetup I owe a lot to AI’s ability to write emails for me, but surprisingly this isn’t what’s meant by technical debt. Francesca does a better job explaining technical debt in ML systems because of their iterative, experimental nature and the absence of time-tested abstractions to prevent boundary erosion. This means you need to devote a lot of energy to fix problems caused by prioritizing speed and ease. She looks at three main sources of technical debt: code duplication, configuration debt, and undeclared consumers. Awareness is crucial in dealing with technical debt, and including configurations in code reviews can help alleviate these problems early on. You’ll only be left with a debt of gratitude after watching this talk. Watch it here [https://home.mlops.community/home/videos/technical-debt-in-ml-systems] [Watch it here](https://home.mlops.community/home/videos/technical-debt-in-ml-systems) ## Why Do We Need A Purpose-Built Database For Multimodal Data? MMDB. No, not the latest thing I use to hallucinate more than ChatGPT. Instead, it's for Multimodal Database, something this blog looks at to help tackle the data management challenges modern AI apps face due to the lack of multimodal data support. As industries like e-commerce, retail, and healthcare handle text, images, audio, and video, existing databases become inefficient and complex. This fragmentation creates silos, bogging down AI performance. A multimodal database tailored for AI can streamline data pipelines by storing and indexing diverse data types in a unified system. This reduces engineering complexity, enhances machine learning workflows, and frees businesses to focus on innovation rather than grappling with data infrastructure. Have a read and expand your mind! With thanks to Vishakha Gupta for their contribution. [Why Do We Need A Purpose-Built Database For Multimodal Data?](https://mlops.community/why-do-we-need-a-purpose-built-database-for-multimodal-data/) ## Yabba Dabba Deploy Amazon launched Bedrock, a managed AWS service for scaling generative AI applications, offering access to leading FMs via a unified API. It supports RAG and model customization within a secure VPC, and integration with AWS services for operational flexibility. Yabba Data Doo [https://arxiv.org/abs/2405.03989] Introducing a method to parse and vectorize semi-structured data, enhancing RAG in LLMs by converting data into .docx for structured extraction and using Pinecone to build vector databases. ABBA Daaba Doo [https://xebia.com/blog/lessons-learned-from-a-diy-llm-benchmark/] Though it’s Eurovision soon it’s not a blog about the band, But about the importance of evals based on your own task, In this case enclosed rhyme poems which are a difficult ask, But once you’ve read the blog you’ll be sure to understand. (Yeah, I can see why AI struggles with it!) [Yabba Dabba Deploy](https://aws.amazon.com/bedrock/studio/) ## IRL Meetups Bristol [https://lu.ma/93ew7j0o] - May 9 (cheers to BJSS 🍻) Denver [https://www.meetup.com/denver-mlops-community/events/300435484/] - May 14 Scotland [https://www.meetup.com/scotland-mlops-community/events/300633800/] - May 16 San Francisco [https://www.meetup.com/mlops-community/events/300785514/] - May 17 (📣 shoutout to Aporia) Amsterdam [https://www.meetup.com/amsterdam-mlops-community/events/300696760/] - May 28 Munich [https://www.meetup.com/munich-mlops-community/events/300673506/] - May 28 Stockholm [https://www.meetup.com/stockholm-mlops-community/events/300881538/] - May 30 [Bristol](https://lu.ma/93ew7j0o) 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/2024-05-09-you-know-quality-when-you-see-it-but-apart-from-the-mirror-w