AI-search webinar puts retrieval ahead of bigger rerankers

The New Stack's Oct. 13 session with Vespa.ai will examine multistage retrieval for RAG systems and agents.

1 Source, 1h ago, first seen 1h ago

TLDR

The New Stack has scheduled an Oct. 13 webinar with Vespa.ai's Bonnie Chase and Jenny Morris on a recurring AI-search problem: a reranker cannot recover relevant material that first-stage retrieval never surfaced. The session will present retrieval as a funnel, beginning with efficient candidate generation, narrowing results through lexical, vector and hybrid methods, and reserving expensive ranking models for the candidates where they can improve the outcome. The advertised agenda also covers latency, compute and relevance tradeoffs, plus the different demands of iterative search by AI agents.

Featured Source

The New Stack has scheduled a live webinar for Oct. 13 on a basic constraint in AI search: a reranker cannot rescue a relevant document or passage that the retrieval stage never included in its candidate set.

The session, set for 10 a.m. Pacific and 1 p.m. Eastern, will feature Bonnie Chase and Jenny Morris of Vespa.ai. Its argument is aimed at retrieval-augmented generation systems and AI agents whose search pipelines can become slower and more expensive as teams compensate for weak first-stage retrieval by sending larger candidate sets through bigger models.

Retrieval as a funnel

The organizers propose treating retrieval and ranking as a sequence rather than one expensive pass. Efficient candidate generation comes first, followed by progressive narrowing through lexical search, vector search and hybrid combinations. More expensive ranking models are then reserved for the smaller group of candidates where deeper scoring can change the result.

That structure makes recall an early design problem. If the right material never enters the candidate set, later ranking quality cannot restore it. If the first stage retrieves too much, downstream inference can become the costliest part of the pipeline.

Agents change the search loop

The advertised agenda also covers how retrieval changes when an AI agent, rather than a person, repeatedly searches, evaluates evidence and decides what to place in context. That loop can multiply latency and compute costs across steps, making the boundary between retrieval and ranking more consequential.

The session is promotional material from The New Stack and Vespa.ai, not an independent benchmark. Its practical focus is on deciding where lexical, vector and hybrid methods belong, and where expensive machine-learning inference improves relevance enough to justify its cost.

AI-search webinar puts retrieval ahead of bigger rerankers

The New Stack's Oct. 13 session with Vespa.ai will examine multistage retrieval for RAG systems and agents.

1 Source, 1h ago, first seen 1h ago

AI-search webinar puts retrieval ahead of bigger rerankers

TLDR

Retrieval as a funnel

Agents change the search loop

AI-search webinar puts retrieval ahead of bigger rerankers

TLDR

Retrieval as a funnel

Agents change the search loop

1 Source

Combined views

Sentiment

Summary

Retrieval as a funnel

Agents change the search loop

Retrieval as a funnel

Agents change the search loop

Combined views

Sentiment

Summary

Sentiment

Summary

1 Source

1 Source

1 Source