Achieving Production-level Performance in RAG with DSPy, Parea, and DVC

RAG PoCs are easy but production-level performance is hard. Jointly optimizing retrieval and generation is a time-consuming process with backtracking the best solution being even harder. What if you could focus solely on defining the architecture of your RAG pipeline? Enter iterating on RAG apps with DSPy, DVC & Parea. In this talk, we’ll explore how self-improving systems can be used to build RAG systems whose retrieval steps course correct and generation steps use optimal few-shot examples. Learn how to leverage DSPy, DVC, and Parea to iterate faster on different RAG approaches. Bio: After completing his M.Sc. in math at NYU on a Fulbright scholarship, Joschka started researching applications of deep learning to medical imaging at Covera Health. After that he built multi-modal semantic search systems and automatic code generation software at Jina AI. Last summer he participated in Y Combinators S23 batch and co-founded Parea AI - a SDK & platform for testing, collaborating & monitoring LLM apps. Link to slides: https://docs.google.com/presentation/d/1kTxeEz909pBYc4n5xUtb25vFPTVAwBys8Mw-qMdKSjE/edit#slide=id.g1a581105611_0_7810 Link to Google Colab notebook: https://colab.research.google.com/drive/1iwHV0iS39Gz6asVofscQQ28oZBwxFh3y?usp=sharing Other content that may be of interest: How to choose a Vector Database by Noé Achace: https://youtube.com/live/aX_hdQEintc?feature=share Great Practices for Retrieval Augmented Generation (RAG) in Production by Noé Achace: https://youtu.be/vZTvzEuOhMk How to create Scalable and Distributed workflows with DVC and Ray by Mikhaila Rozhkov: https://youtube.com/live/3QvuQ4BC-Q8?feature=share