The Real E2E RAG Stack // Sam Bean, Rewind AI // #217

Thank you to Zilliz our wonderful sponsors of this episode create some amazing stuff with Zilliz RAG - https://zilliz.com/vector-database-use-cases/llm-retrieval-augmented-generation Sam Bean is a seasoned AI and machine learning expert, specializing in Large Language Models (LLMs) and search tech. With a computer science background and a drive for innovation, Sam leads the team at Rewind AI in leveraging advanced tech to tackle complex challenges. MLOps podcast #217 with Sam Bean, Software Engineer (Applied AI) at Rewind.ai, The Real E2E RAG Stack. // Abstract What does a fully operational LLM + Search stack look like when you're running your own retrieval and inference infrastructure? What does the flywheel really mean for RAG applications? How do you maintain the quality of your responses? How do you prune/dedupe documents to maintain your document quality? // Bio Sam has been training, evaluating, and deploying production-grade inference solutions for language models for the past 2 years at You.com. Previous to that he built personalization algorithms at StockX. // MLOps Jobs board https://mlops.pallet.xyz/jobs // MLOps Swag/Merch https://mlops-community.myshopify.com/ // Related Links Website: https://github.com/sam-h-bean/ REinforced Self Training (REST) - https://arxiv.org/pdf/2308.08998.pdf REST meets REACT - https://arxiv.org/pdf/2312.10003.pdf --------------- ✌️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 Sam on LinkedIn: https://www.linkedin.com/in/samuel-h-bean/ Timestamps: [00:00] Sam's preferred coffee [00:11] Takeaways [03:52] A competitive coding pinball player [07:18] Sam's MLOps journey [10:33] Search Challenges with ML [15:04] Expensive evaluation [21:04] Labeling Parties Boost Data Quality [24:10] Zeno's Paradox of Motion [25:51] Sam's job at Rewind AI [29:35] Multimodal RAG [30:59 - 32:06] Zilliz Ad [32:07] University of Prague paper leak [36:38] Signals behind the scenes [39:28] Content Over Metadata Approach [43:22] Optionality around evaluation and search [48:35] Incremental Robustness Building [51:33] Solid Foundations for Success [53:42] Production RAGs [1:00:06] Thoughts on DSPy [1:05:40] Using DSPy in Production [1:08:26] Wrap up