Practical Synthetic Data Strategies for RAG/Agentic Systems

Explore practical applications of synthetic data in developing RAG (Retrieval-Augmented Generation) and agentic systems through this 32-minute conference talk from the Linux Foundation's Edge US event. Discover how synthetic data, which has been instrumental in recent LLM progress and post-training, can be effectively applied beyond traditional use cases to enhance context engineering development for RAG and agent systems using open-source tools. Learn how RAG systems combine search and recommender system methods for context retrieval, and understand how synthetic data addresses the gaps between evaluation and deployment data by maintaining distributional relationships from real data through data invariance while adding diversity. Examine how synthetic data enables smoother regression testing of query retrieval mechanisms by generating diverse query-answer pairs, accelerating testing cycles and facilitating improvements in prompt optimization and fine-tuning. Understand the role of synthetic data in RAG system development through faster iterative feedback loops, with practical demonstrations of leveraging existing open-source tools for synthetic data generation. Gain insights into future directions for synthetic data applications in AI system development, presented by Kevin Noel from Uzabase at this premier vendor-neutral open source conference focused on collaboration, knowledge sharing, and exploring the latest innovations in open source technology.