Optimizing Retrieval for Agentic Systems
summary
Workshop claims that retrieval moves past naive RAG into agentic search / self-RAG with LLM-driven query formulation and tool selection, demonstrates code labs (Docker Jupyter notebooks, GitHub) and Vespa OpenSearch/Solr integrations, and shows measurable wins: late interaction (ColBERT/ColPali) outperforms single-vector models and can replace OCR for PDFs/images, reflected intelligence feedback loop captures behavioral signals for personalization, and a semantic knowledge graph approach boosts query rewriting; course material and live labs back the results.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 7:35RAG to agentic search
- 15:01Search relevance problem
- 32:00Modern retrieval techniques
- 39:29Reflected intelligence models
- 57:29Multimodal and hybrid search
- 1:06:37Late interaction retrieval
- 1:15:47Semantic knowledge graphs
- 1:28:08Course resources
related talks
40mAgentic Retrieval: Building Self-Optimizing Search Systems
22mOrchestrating Multimodal RAG With Agentic Workflows With OpenSearch
40mGo Beyond Basic RAG with Agentic Behavior
50mRetrieval Augmented Generation Beyond Toy Demos
51mDeveloping a Custom RAG System
01h 28mFrom RAG to Riches Building Enterprise Search Systems with LLMs
31mOrchestrating Multimodal RAG With Agentic Workflows
28mNavigating Challenges and Enhancing Performance of LLM based Applications