--- title: "Producing the goods for production-ready AI." newsletter: "MLOps Community" date: 2024-06-20 source: https://aaif.live/newsletters/mlopscommunity/2024-06-20-producing-the-goods-for-production-ready-ai --- # Producing the goods for production-ready AI. *Plus, synth-sational, vector victories, multimodal RAG in the bag, find those insights quickly, and hidden gems.* *MLOps Community — Agentic AI Foundation, 2024-06-20* ## How to Build Production-Ready AI Models for Manufacturing // LatticeFlow Roundtable ## MLOps Community Roundtable Manufacturing is difficult—just ask Willy Wonka. Adding AI is an extra challenge. Thankfully, this roundtable explores the challenges and strategies involved in deploying AI in manufacturing, particularly in computer vision and generative AI. It covers training models to recognize the ideal state of components rather than all possible defects and the creation of carbon digital twins for process optimization and carbon emission control. Practical applications like battery inspection and surface and dimension inspection are also discussed. Watch this, then train the model that writes the songs about kids stuck in machines. Video [https://home.mlops.community/home/videos/how-to-build-production-ready-ai-models-for-manufacturing-exclusive-latticeflow-roundtable] || Spotify [https://open.spotify.com/episode/6jMDzPPBSZjK39Et9CfPy5?si=QvAOBQgtQZKtSgh4tKG-fw] || Apple [https://podcasts.apple.com/us/podcast/how-to-build-production-ready-ai-models-for-manufacturing/id1505372978?i=1000659037847] ## The Power of Synthetic Data and Agent Evaluations // Boris Selitser // MLOps Podcast #241 Important note: Synth data shows preferences for a DX7 over a Prophet-5. Synthetic data is something else. Helpfully, this episode looks at the latter. Boris walked me through its importance for realistic testing scenarios and application-specific metrics for assessment, emphasizing modular, microservice-like agent design patterns to enhance reliability and control. He also highlighted the challenges in deploying AI agents, the evolving AI framework landscape, and the necessity of iterative processes, tailored data, and strategic metrics. After listening, synthetic data will be music to your ears. Video [https://home.mlops.community/home/videos/navigating-the-ai-frontier-the-power-of-synthetic-data-and-agent-evaluations-in-llm-development] || Spotify [https://open.spotify.com/episode/34Lx6gllnP2ySpYqXIvhiW?si=6MuJzcLUSFS4rfCBZlEBMg] || Apple [https://podcasts.apple.com/us/podcast/navigating-the-ai-frontier-the-power-of-synthetic/id1505372978?i=1000659407352] [https://podcasts.apple.com/us/podcast/navigating-the-ai-frontier-the-power-of-synthetic/id1505372978?i=1000659407352](https://podcasts.apple.com/us/podcast/navigating-the-ai-frontier-the-power-of-synthetic/id1505372978?i=1000659407352) [https://home.mlops.community/home/videos/navigating-the-ai-frontier-the-power-of-synthetic-data-and-agent-evaluations-in-llm-development](https://home.mlops.community/home/videos/navigating-the-ai-frontier-the-power-of-synthetic-data-and-agent-evaluations-in-llm-development) [https://open.spotify.com/episode/34Lx6gllnP2ySpYqXIvhiW?si=6MuJzcLUSFS4rfCBZlEBMg](https://open.spotify.com/episode/34Lx6gllnP2ySpYqXIvhiW?si=6MuJzcLUSFS4rfCBZlEBMg) ## Job of the Week [https://www.linkedin.com/jobs/view/3915458222/?eBP=NOT_ELIGIBLE_FOR_CHARGING&refId=DmwThV6uyuW1EO%2BgU6ckkg%3D%3D&trackingId=2oViX04wNGIq7Voh5GCayA%3D%3D&trk=flagship3_search_srp_jobs](https://www.linkedin.com/jobs/view/3915458222/?eBP=NOT_ELIGIBLE_FOR_CHARGING&refId=DmwThV6uyuW1EO%2BgU6ckkg%3D%3D&trackingId=2oViX04wNGIq7Voh5GCayA%3D%3D&trk=flagship3_search_srp_jobs) ## Fresh Data, Smart Retrieval with Milvus & Jina CLIP // MLOps Mini Summit #7 ## MLOps Community Mini Summit Fresh Data, Smart Retrieval with Milvus & Jina CLIP // MLOps Mini Summit #7 Vectors making you volatile? RAG getting you in a rage? Find your balance with this mini-summit on techniques for maintaining fresh data in vector databases, focusing on RAG systems. We'll start with Milvus, an open-source vector database, covering its architecture, features, and benefits. Learn strategies for data insertion, updating, and management to keep your data current. Next, explore Gina Clip, a multi-modal embedding, including its training process, ability to handle long texts, and practical application in vector search using DocArray. View for vector victories and reap RAG rewards! Watch it here [https://home.mlops.community/home/videos/fresh-data-smart-retrieval-milvus-and-jina-clip-explained] ## Building Multimodal RAG // Hamza Farooq & Darshil Modi // MLOps IRL Meetup #82 Silicon Valley ## MLOps Community IRL Meetup Building Multimodal RAG // Hamza Farooq & Darshil Modi // MLOps IRL Meetup #82 Silicon Valley Multimodal data can mean PDF stands for Problematic Data Formats. This IRL talk examines a RAG architecture designed to address the challenges of processing multimodal data from PDFs, incorporating text, images, diagrams, and handwritten notes. Traditional data chunking methods don’t always preserve context, but this method optimally separates and processes different data types. Text segments are enriched with metadata to enhance retrieval, while advanced vision models like LAVA and GPT interpret visual content. This enables high-precision retrieval from a searchable vector database, improving data extraction for sectors like finance. Click and watch to make PDF mean Poly-Data Fusion. Watch it here [https://home.mlops.community/home/videos/building-multimodal-rag] [Watch it here](https://home.mlops.community/home/videos/building-multimodal-rag) ## Semantic Search to Glean Valuable Insights from Podcasts With so many great podcasts sharing so much knowledge, it can be hard to pinpoint the specific insight you’re after. Thankfully, this blog by a fellow podcaster outlines a project for transcribing podcast audio files using OpenAI's Whisper model and storing the results in ApertureDB for easy querying. The process includes setting up a transcription environment in Google Colab, converting audio to text, and managing metadata in ApertureDB. The guide covers installing the necessary libraries, running the Whisper model for transcription, and ingesting the text files into the vector database step by step. You’ll notice I didn’t specifically mention our podcast in the intro, but it was nice that you thought of it. With thanks to Sonam Gupta for their contribution. [Semantic Search to Glean Valuable Insights from Podcasts](https://home.mlops.community/home/blogs/semantic-search-to-glean-valuable-insights-from-podcasts-part-1) ## Hidden Gems ## IRL Meetups San Francisco [https://www.aiqualityconference.com/] - June 25 - 🙏with Kolena Denver [https://mlops.community/event/pizza-pints-expert-speaker-intro-to-machine-learning-toolkits-with-kubeflow/] - July 30 [San Francisco](https://www.aiqualityconference.com/) 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/2024-06-20-producing-the-goods-for-production-ready-ai