MLOps.community

Karl Fezer, Intelligence & MLOps expert, discusses biases, defining intelligence, and the future of large language models in AI. He emphasizes the importance of efficient high-impact tasks in MLOps. The conversation touches on philosophical tangents but relates back to practical applications of these concepts.

MLOps Coffee Sessions #147 with Alex DeBrie, Something About Databases co-hosted by Abi Aryan. // Abstract For databases, it feels like we're in the middle of a big shift. The first 10-15 years of the cloud were mostly about using the same core infrastructure patterns but in the cloud (SQL Server, MySQL, Postgres, Redis, Elasticsearch).   In the last few years, we're finally seeing data infrastructure that is truly built for the cloud. Elastic, scalable, resilient, managed, etc. Early examples were Snowflake + DynamoDB. The most recent ones are all the 'NewSQL' contenders (Cockroach, Yugabyte, Spanner) or the 'serverless' ones (Neon, Planetscale). Also seeing improvements in caching, search, etc. Exciting times! // Bio Alex is an AWS Data Hero and self-employed AWS consultant and trainer. He is the author of The DynamoDB Book, a comprehensive guide to data modeling with DynamoDB. Previously, he worked for Stedi and for Serverless, Inc., creators of the Serverless Framework. He loves being involved in the AWS & serverless community, and he lives in Omaha, NE with his family. // MLOps Jobs board   https://mlops.pallet.xyz/jobs // MLOps Swag/Merch https://mlops-community.myshopify.com/ // Related Links Key Takeaways from the DynamoDB Paper: https://www.alexdebrie.com/posts/dynamodb-paper/ Understanding Eventual Consistency in DynamoDB: https://www.alexdebrie.com/posts/dynamodb-eventual-consistency/ Two Scoops of Django 1.11: Best Practices for the Django Web Framework: https://www.amazon.com/Two-Scoops-Django-1-11-Practices/dp/0692915729CAP or no CAP? Understanding when the CAP theorem applies and what it means: https://www.alexdebrie.com/posts/when-does-cap-theorem-apply/ Stop fighting your database/ The DynamoDB book: https://dynamodbbook.com/ --------------- ✌️Connect With Us ✌️ ------------- Join our slack community: https://go.mlops.community/slack Follow us on Twitter: @mlopscommunity Sign up for the next meetup: https://go.mlops.community/register Catch all episodes, blogs, newsletters, and more: https://mlops.community/ Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/ Connect with Abi on LinkedIn: https://www.linkedin.com/in/abiaryan/ Connect with Alex on LinkedIn: https://www.linkedin.com/in/alex-debrie/ Timestamps: [00:00] Alex's preferred coffee [00:27] Introduction to Alex DeBrie and DynamoDB [01:05] Takeaways [03:47] Please write down your reviews and you might have a book of Alex! [04:57] Alex's journey from being an Attorney to becoming a Data Engineer [07:31] Why engineering? [10:07] Serverless Data [12:54] Before Airflow [15:41] Batch vs streaming [17:03] Difficulties in Batch and Streaming side [19:45] Modern data infrastructure databases [24:37] Cloud Ecosystem Maturity [27:59] New generation type of Snowflake [29:47] Comparing databases [30:58] What's next on connectors from 2 perspectives? [34:25] Management services at the MLOps level [36:38] DynamoDB [39:32] Why do you like DynamoDB? [41:00] Data used in DynamoDB and size limits [43:46] Comparison of tradeoffs between DynamoDB and Redis [45:52] Preferred opinionated databases [48:43] CAP or no CAP? Understanding when the CAP theorem applies and what it means [52:10] The DynamoDB book [56:17] Chapter you want to expand on the book [57:43] Next book to write [59:25] ChatGPT iterations [1:01:59] Data modeling book wished to be written [1:03:27] Wrap up

MLOps Coffee Sessions #146 with Shalabh Chaudri, The Ops in MLOps - Process and People co-hosted by Abi Aryan. // Abstract Shalabh talks through their newfound appreciation for the MLOps perspective from a customer success standpoint. Shalabh's emphasis on setting realistic expectations and ensuring the delivery of promised value adds is particularly valuable.     Generally, this episode provides a unique and insightful perspective on MLOps from the lens of customer success. // Bio Shalabh has worked in the MLOps domain since 2020 at Algorithmia and Union AI. His experience spans startups and small and large public companies. He has 10+ years of experience in the design, delivery, adoption, and business value realization of B2B infrastructure and platform solutions. // MLOps Jobs board   https://mlops.pallet.xyz/jobs // MLOps Swag/Merch https://mlops-community.myshopify.com/ // Related Links https://www.union.ai/ --------------- ✌️Connect With Us ✌️ ------------- Join our slack community: https://go.mlops.community/slack Follow us on Twitter: @mlopscommunity Sign up for the next meetup: https://go.mlops.community/register Catch all episodes, blogs, newsletters, and more: https://mlops.community/ Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/ Connect with Abi on LinkedIn: https://www.linkedin.com/in/abiaryan/ Connect with Shalabh on LinkedIn: https://www.linkedin.com/in/shalabhchaudhri/ Timestamps: [00:00] Shalabh's preferred coffee [01:18] Takeaways [02:57] Huge shout out to Union AI! [03:46] Reviews [05:26] Shalab's journey [07:00] The people and process of MLOps [10:25] Accuracy measures and Multiple Stakeholders [13:01] UnionAI's success where others fall short [14:45] Legacy equipment [17:06] Legacy tools versus open source [19:27] Cataloging solution [22:51] Stakeholders and maturity levels [24:26] People and Process in MLOps [29:00] Collaboration for Machine Learning [31:08] Overcoming challenges [34:17] AI and leadership decision-making [35:33] Legacy Companies and AI [39:39] Common pitfalls   [42:24] Neglecting ROI [46:25] Speaking to each level [49:50] Being realistic [51:29] Becoming a champion [53:08] Transitioning to machine learning [55:25] Customer's Skill and Success needed in ML   [57:46] Different sizes of companies 

MLOps Coffee Sessions #145 with Sahil Khanna, Griffin, ML platform at Instacart co-hosted by Mike Del Balso. // Abstract The conversation revolves around the journey of Instacart in implementing machine learning, starting from batch processing to real-time processing. The speaker highlights the importance of real-time processing for businesses and the relevance of Instacart's journey to other machine learning teams.    Sahil emphasizes the soft factors, such as staying customer-focused and the right approach, that contributed to the success of Instacart's machine learning implementation. We also recommend two blog posts by Sahil about Instacart's journey. // Bio Sahil is currently a machine learning engineer at Instacart, where they are building a centralized platform for the training, deployment, and management of diverse ML applications. Before Instacart, Sahil developed ML training and inference platforms at Etsy. // MLOps Jobs board   https://mlops.pallet.xyz/jobs // MLOps Swag/Merch https://mlops-community.myshopify.com/ // Related Links --------------- ✌️Connect With Us ✌️ ------------- Join our slack community: https://go.mlops.community/slack Follow us on Twitter: @mlopscommunity Sign up for the next meetup: https://go.mlops.community/register Catch all episodes, blogs, newsletters, and more: https://mlops.community/ Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/ Connect with Mike on LinkedIn: https://www.linkedin.com/in/michaeldelbalso/ Connect with Sahil on LinkedIn: www.linkedin.com/in/sahil-khanna-umd

MLOps Coffee Sessions #144 with Matthew Dombrowski, Non-traditional Career Paths in MLOps co-hosted by Mihail Eric. // Abstract Let's explore the different aspects of ML and data roles and the variety of responsibilities each role entails! This conversation emphasizes the need for understanding the unique insights each role provides and the similarities in responsibilities and soft skills that are required across different roles.    This episode also highlights the significance of stakeholder alignment in the context of working in big companies and the importance of navigating these complexities for a successful career in ML. // Bio Matt has performed a number of MLOps positions including Solutions Consultant, Solutions Architect, and Product Manager from startups to large organizations. In his current role, Matt builds tools to help social media influencers discover unique and exciting Amazon products to recommend to their audiences. // MLOps Jobs board   https://mlops.pallet.xyz/jobs // MLOps Swag/Merch https://mlops-community.myshopify.com/ // Related Links --------------- ✌️Connect With Us ✌️ ------------- Join our slack community: https://go.mlops.community/slack Follow us on Twitter: @mlopscommunity Sign up for the next meetup: https://go.mlops.community/register Catch all episodes, blogs, newsletters, and more: https://mlops.community/ Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/ Connect with Mihail on LinkedIn: https://www.linkedin.com/in/mihaileric/ Connect with Matt on LinkedIn: https://www.linkedin.com/in/matthewdombrowski/ Timestamps: [00:00] Matt's preferred coffee [00:28] Mihail's new creation [05:09] Introduction to Matthew Dombrowski [06:02] Takeaways [08:30] Pizza and coffee nerds [10:54] Data careers [13:35] Matt's progression through the ml sphere [20:10] Dealing with machine learning [23:20] Transition from deep technical implementer to PM role [27:42] Data is a product [29:30] From start-ups to big companies [32:41] Ambiguity of ML [36:17] Matt's daily routine [40:23] Social media influencers [42:07] Developer advocate [44:00] Stakeholder alignment [49:41] Non-traditional career paths military influence [54:11] Good ways to recommend people to get into ML [57:56] MLOps Meetups all over the world [59:00] Wrap up

MLOps Coffee Sessions #143 with Jill Chase & Manmeet Gujral, Investing in the Next Generation of AI & ML. // Abstract Investors are currently focusing on developer tooling and the foundational AI model movement, as they have seen explosive growth in this area. This podcast explores the impact of foundational models on investment thesis and the future of machine learning operations. The discussion also touches on the idea of generative AI and large language models, and their potential impact on MLOps in the next 10 years. Jill and Manmeet from Capital G share their insights on this topic. // Bio Jill Chase Jill is an investor at CapitalG where she focuses on enterprise software, with an emphasis on data infrastructure and AI/ML. Prior to joining CapitalG, Jill worked in senior startup operating roles, both as the CEO of a private equity-backed business and as the founder of a Y Combinator-backed startup. Jill graduated magna cum laude from Williams College with a dual degree in Economics and Psychology and was captain of the women’s basketball team. She came out to the West Coast to earn an MBA from the Stanford Graduate School of Business, but she was born and raised in Boston where she had the opportunity to cheer on the most impressive era of professional sports a city has ever experienced (Go Patriots). She lives in the Bay Area with her husband where they spend weekends doing as many outside activities as possible, such as pickleball, tennis, hiking, and running. Manmeet Gujral Manmeet is a member of the CapitalG investment team where he focuses on enterprise software, AI & ML, open source, and product-led-growth companies. Prior to joining CapitalG in 2021, Manmeet worked in product marketing and operations at Tecton. Before that, he worked as a consultant at Bain & Company in San Francisco where he specialized in the go-to-market strategy for technology companies and private equity investment diligence. Manmeet is originally from Albany, New York, and graduated from Dartmouth College with a dual degree in Computer Science and Economics. Manmeet is highly opinionated about pizza, an avid New York sports fan, and always willing to share his latest house or hip-hop playlists. // MLOps Jobs board   https://mlops.pallet.xyz/jobs // MLOps Swag/Merch https://mlops-community.myshopify.com/ // Related Links --------------- ✌️Connect With Us ✌️ ------------- Join our slack community: https://go.mlops.community/slack Follow us on Twitter: @mlopscommunity Sign up for the next meetup: https://go.mlops.community/register Catch all episodes, blogs, newsletters, and more: https://mlops.community/ Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/ Connect with Jill on LinkedIn: https://www.linkedin.com/in/jill-greenberg-chase-53747538/ Connect with Manmeet on LinkedIn: https://www.linkedin.com/in/manmeet-gujral/ Timestamps: [00:00] Manmeet and Jill's preferred coffee [00:25] Takeaways [01:31] CapitalG, Jill and Manmeet's Background [05:12] Sideswiping MLOps by Foundational Models [08:50] MLOps space and the market revenue   [14:50] Foundational models B to C style [20:37] Foundational models taking over [23:00] Uninnovative sentiments [27:50] 2 prototypes of companies [31:51] Finding product market fit [36:20] MLOps market growth changes [40:30] Monster valuations [41:43] The ones that got away [44:07] Wrap up

MLOps Coffee Sessions #142 with Murtuza Shergadwala, Approaches to Fairness and XAI co-hosted by Abi Aryan. This episode is sponsored by Fiddler AI. // Abstract The field of Explainable Artificial Intelligence (XAI) is continuously evolving, with an increasing focus on providing model-centric explanations in a human-centric manner. However, better frameworks and training for users are needed to fully utilize the potential of XAI tools.   Additionally, there is a discrepancy in the approach to fairness in XAI, with the industry approaching it from a regulatory standpoint, while academia is engaging in more discussion and research on the topic. // Bio Dr. Murtuza Shergadwala is a data scientist at Fiddler AI. His background is in human-machine interaction and design decision-making. He received his Ph.D. from Purdue University in Mechanical Engineering. Prior to Fiddler, he was a postdoc at the Games User Interaction and Intelligence Lab at UC Santa Cruz where he focused on using Bayesian approaches for modeling cognition and investigating the theory of mind. He’s super passionate about fairness in AI for underrepresented communities. // MLOps Jobs board   https://mlops.pallet.xyz/jobs // MLOps Swag/Merch https://mlops-community.myshopify.com/ // Related Links https://murtuzashergadwala.wixsite.com/murtuza https://www.fiddler.ai/blog/detecting-intersectional-unfairness-in-ai-part-1 --------------- ✌️Connect With Us ✌️ ------------- Join our slack community: https://go.mlops.community/slack Follow us on Twitter: @mlopscommunity Sign up for the next meetup: https://go.mlops.community/register Catch all episodes, blogs, newsletters, and more: https://mlops.community/ Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/ Connect with Abi on LinkedIn: https://www.linkedin.com/in/abiaryan/ Connect with Murtuza on LinkedIn: https://www.linkedin.com/in/murtuza-shergadwala/ Timestamps: [00:00] Moto's preferred coffee [00:35] Introduction to Murtuza Shergadwala [01:06] Takeaways [04:30] Huge shout out to Fiddler AI for sponsoring this episode! [05:00] Don't forget to like, comment, and subscribe. Give us a rating   [06:10] Moto's background and transition to Human-centric AI [10:52] Decision-making behaviors of engineering designers in design contests [15:10] Gaining insights from data decisions [18:00] Defining latent variables [20:32] Designer's perspective on building systems [23:14] XAI as a movement [27:47] Selling regulations and bridging the gap [32:18] Data integrity towards detecting outliers alerting and data drifts [34:32] Dealing with alerts and alert fatigue [37:31] Approaches and their limitations [39:10] Alert-level systems [42:19] Alerts putting into practice [45:30] Creative alerts [47:02] One solution fits all? [50:08] Wrap up

MLOps Coffee Sessions #141 with Stephen Bailey, Airflow Sucks for MLOps co-hosted by Joe Reis. // Abstract Stephen discusses his experience working with data platforms, particularly the challenges of training and sharing knowledge among different stakeholders. This talk highlights the importance of having clear priorities and a sense of practicality and mentions the use of modular job design and data classification to make it easier for end users to understand which data to use.    Stephen also mentions the importance of being able to move quickly and not getting bogged down in the quest for perfection. We recommend Stephen's blog post "Airflow's Problem" for further reading. // Bio Stephen has worked as a data scientist, analyst, manager, and engineer, and loves all the domains equally. He currently works at Whatnot, a collectibles marketplace that focuses on live shopping, and has previously worked in privacy tech at Immuta. He has his Ph.D. from Vanderbilt University in educational cognitive neuroscience, but it has yet to help him understand why his three children are so crazy. // MLOps Jobs board   https://mlops.pallet.xyz/jobs // MLOps Swag/Merch https://mlops-community.myshopify.com/ // Related Links Airflow's Problem blog post: https://stkbailey.substack.com/p/airflows-problem Airflow's Problem and the reception it got on Hacker News: https://news.ycombinator.com/item?id=32317558 --------------- ✌️Connect With Us ✌️ ------------- Join our slack community: https://go.mlops.community/slack Follow us on Twitter: @mlopscommunity Sign up for the next meetup: https://go.mlops.community/register Catch all episodes, blogs, newsletters, and more: https://mlops.community/ Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/ Connect with Joe on LinkedIn: https://www.linkedin.com/in/josephreis/ Connect with Stephen on LinkedIn: https://www.linkedin.com/in/stkbailey/ Timestamps: [00:00] Stephen's preferred coffee [00:19] Introduction to co-host Joe Reis [01:40] Takeaways [06:29] Subscribe to our newsletters! [06:55] Shout out to our sponsor, Wallaroo! [08:05] Whatnot [10:47] Stephen's side hustle [14:35] Stephen's work breakdown at Whatnot [18:03] Fundamental tensions in the data world [21:27] Initial questions to answer that you were on the right path [24:06] Recommender systems [28:15] Coordinating with ML teams [29:43] Daxter [31:38] Too advanced, more challenging [34:37] Orchestration layer [36:14] Decision criteria [39:23] Human design aspect of Daxter [40:53] Orchestration layer centralization and sharing knowledge with stakeholders [46:18] Airflow's Problem and the reception it got on Hacker News [51:00] Wrap up

MLOps Coffee Sessions #140 with Sakib Dadi, The Evolution of ML Infrastructure sponsored by Wallaroo. // Abstract The toolkit and infrastructure empowering machine learning practitioners are advancing as ML adoption accelerates. We'll go through the current landscape of ML tooling, startups, and new projects from an investor's perspective. // Bio Sakib is a vice president in the San Francisco office where he primarily focuses on early-stage investments in developer platforms, data infrastructure, and machine learning. He has been involved with Bessemer’s investments in Prefect, Coiled, Arcion, Periscope Data (acquired by Sisense), Okera, npm (acquired by GitHub), and LaunchDarkly. Before joining Bessemer, Sakib worked in product at Viagogo, an international marketplace for buying and selling tickets for live events.      Sakib also worked in the technology investment banking group at Morgan Stanley and as an engineer at Innova Dynamics (acquired by TPK), a startup manufacturing flexible touchscreen displays. // MLOps Jobs board   https://mlops.pallet.xyz/jobs // MLOps Swag/Merch https://mlops-community.myshopify.com/ // Related Links --------------- ✌️Connect With Us ✌️ ------------- Join our slack community: https://go.mlops.community/slack Follow us on Twitter: @mlopscommunity Sign up for the next meetup: https://go.mlops.community/register Catch all episodes, blogs, newsletters, and more: https://mlops.community/   Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/ Connect with Sakib on LinkedIn: https://www.linkedin.com/in/sakib-dadi-77938937/ Timestamps: [00:00] Sakib's preferred coffee [00:13] Introduction to Sakib Dadi [01:33] Sakib's background [02:40] Shout out to this episode's Sponsor, Wallaroo! [04:17] ML investing [05:57] Investing regrets [08:06] Transformers are today what would be tomorrow? [09:18] Company that you wish existed now [10:23] Current thoughts on the MLOps market [12:32] MLOps transition to Generative AI [15:52] Mind maps [17:03] Jasper [22:14] Intersection   [24:10] Differences in models in-house [26:08] Orchestration space [28:23] Nuances of Monitoring [29:20] Demetrios' theory on Monitoring [31:48] Non-funded Monitoring Companies [34:29] Investment risks [36:55] Orchestration markets [39:38] MLOps market at a plateau [42:14] Vertical problems, vertical solutions   [45:45] Sakib starting a company [50:10] Structuring deals [51:50] Infrastructure tools [53:26] Firing a founder   [53:49] Parting ways with a founder [57:07] Wrap up

MLOps Coffee Sessions #139 with Alex Ratner, Putting Foundation Models to Use for the Enterprise co-hosted by Abi Aryan sponsored by Snorkel AI. // Abstract Foundation models are rightfully being compared to other game-changing industrial advances like steam engines or electric motors. They’re core to the transition of AI from a bespoke, less predictable science to an industrialized, democratized practice. Before they can achieve this impact, however, we need to bridge the cost, quality, and control gaps.    Snorkel Flow Foundation Model Suite is the fastest way for AI/ML teams to put foundation models to use. For some projects, this means fine-tuning a foundation model for production dramatically faster by creating programmatically labeling training data. For others, the optimal solution will be using Snorkel Flow’s distill, combine, and correct approach to extract the most relevant knowledge from foundation models and encode that value into the right-sized models for your use case.    AI/ML teams can determine which Foundation Model Suite capabilities to use (and in what combination) to optimize for cost, quality, and control using Snorkel Flow’s integrated workflow for programmatic labeling, model training, and rapid-guided iteration. // Bio Alex Ratner is the Co-founder and CEO of Snorkel AI and an Assistant Professor of Computer Science at the University of Washington.    Prior to Snorkel AI and UW, he completed his Ph.D. in CS advised by Christopher Ré at Stanford, where he started and led the Snorkel open-source project, and where his research focused on applying data management and statistical learning techniques to emerging machine learning workflows such as creating and managing training data and applying this to real-world problems in medicine, knowledge base construction, and more. Previously, he earned his A.B. in Physics from Harvard University. // MLOps Jobs board   https://mlops.pallet.xyz/jobs // MLOps Swag/Merch https://mlops-community.myshopify.com/ // Related Links Website: www.snorkel.ai Huge “foundation models” are turbo-charging AI progress: https://www.economist.com/interactive/briefing/2022/06/11/huge-foundation-models-are-turbo-charging-ai-progress Nemo: Guiding and Contextualizing Weak Supervision for Interactive Data Programming: https://arxiv.org/abs/2203.01382 The Principles of Data-Centric AI Development: https://snorkel.ai/principles-of-data-centric-ai-development/ --------------- ✌️Connect With Us ✌️ ------------- Join our slack community: https://go.mlops.community/slack Follow us on Twitter: @mlopscommunity Sign up for the next meetup: https://go.mlops.community/register Catch all episodes, blogs, newsletters, and more: https://mlops.community/ Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/ Connect with Abi on LinkedIn: https://www.linkedin.com/in/abiaryan/ Connect with Alex on LinkedIn: https://www.linkedin.com/in/alexander-ratner-038ba239/ Timestamps: [00:00] Alex's preferred coffee [01:20] Introduction to Alex Ratner [02:34] Takeaways [04:04] Huge shoutout to our Sponsor, Snorkel AI! [04:39] Comment, rate us, and share our podcasts with your friends!  [04:50] Transfer Learning / Active Learning [11:30] Labeling Heuristics paper on Nemo [18:14] Geocentric AI [21:48] Enterprise use cases on Foundational Models [32:45] Foundational Models into the different Google products [38:36] Progress in Foundational Models [43:55] AutoML Models Baseline Accuracy [44:40] Hosting Infrastructure Snorkel Float vs GCP [46:53] Chris Re's venture capital firm / incubator / machine [51:00] Wrap