Fixing Broken Retrieval: Building Self-Correcting Agent Loops

About Event Most retrieval agents fail quietly. The agent pulls something plausible, generates an answer, and only later do you realize the context was incomplete or wrong. In this hands-on workshop, you'll build and evaluate an agent that recognizes when its retrieval step isn't good enough. Starting from a prebuilt Qdrant-powered agent, you'll run evals, inspect traces, identify failure modes, and add in-loop checks that help the agent decide whether to rewrite the query, search again, or stop. We'll cover practical retrieval-quality signals: low top-result confidence, tightly clustered scores, weak rank gaps, and disagreement between keyword and vector retrieval. Then we'll use those signals to improve the loop and measure whether the changes actually improve answer quality, retrieval…