Retrieval-Augmented Generation (RAG), recall-first RAG, retrieval optimization, semantic search, vector search and production RAG architecture all depend on one critical question: are you retrieving all the relevant information?
In this session, Grzegorz Jurdziński, Senior ML Engineer, explains why standard Top-K retrieval often becomes the bottleneck in production RAG systems, and what alternatives exist when recall matters more than latency.
The webinar covers:
- why Top-K retrieval can miss critical evidence
- recall vs precision trade-offs
- similarity thresholds and reranking
- using LLMs as relevance judges
- retrieval evaluation strategies
- cost vs recall optimization
- choosing the right retrieval strategy for enterprise AI
Timeline
00:00 What recall means
01:28 Why Top-K fails
03:32 Better retrieval strategies
07:02 LLM reranking
09:57 Agentic retrieval
12:32 Choosing the right approach
Speaker
Grzegorz Jurdziński
Senior ML Engineer
