The Real End-to-End RAG Stack - Lecture 217

Dive into a comprehensive podcast episode featuring Sam Bean, Software Engineer in Applied AI at Rewind.ai, discussing the intricacies of the real end-to-end Retrieval-Augmented Generation (RAG) stack. Explore the complexities of running retrieval and inference infrastructure for Language Model (LLM) applications, the concept of the RAG flywheel, methods for maintaining response quality, and techniques for pruning and deduplicating documents. Gain insights into Sam's experience in training, evaluating, and deploying production-grade inference solutions for language models, as well as his background in personalization algorithms. Learn about cutting-edge concepts like REinforced Self Training (REST) and its integration with REACT. Discover practical approaches to search challenges, expensive evaluation processes, and the importance of data quality in machine learning operations. Delve into discussions on multimodal RAG, content-focused approaches, and the implementation of DSPy in production environments. This episode offers valuable knowledge for professionals and enthusiasts in the fields of artificial intelligence, machine learning, and natural language processing.

[] Sam's preferred coffee
[] Takeaways
[] A competitive coding pinball player
[] Sam's MLOps journey
[] Search Challenges with ML
[] Expensive evaluation
[] Labeling Parties Boost Data Quality
[] Zeno's Paradox of Motion
[] Sam's job at Rewind AI
[] Multimodal RAG
[30:59 - ] Zilliz Ad
[] University of Prague paper leak
[] Signals behind the scenes
[] Content Over Metadata Approach
[] Optionality around evaluation and search
[] Incremental Robustness Building
[] Solid Foundations for Success
[] Production RAGs
[] Thoughts on DSPy
[] Using DSPy in Production
[] Wrap up