Multi-vector search: how to choose among single-vector bi-encoders, late interaction (ColBERT-family), and cascaded reranking—grounded in the Weaviate podcast with LightOn’s Amélie Chatelain and Antoine Chaffin.
Multi-stage language programs and automatic prompt optimization: from DSPy to MIPRO—proposal, bootstrapping, and combinatorial search; credit assignment; meta-proposers; and how they relate to RAG, agents, and fine-tuning.
Judge-time compute: stacking structured, composable weak-model calls at evaluation time instead of assuming one expensive judge pass is enough—Verdict, agreement metrics, and production guardrails, with evidence boundaries called out.
When JEP 395 bundles immutable carriers, nominal tuples, and record patterns, any evolution beyond its constraints loses both compact syntax and expressive power on the pattern-matching side. Project Amber is elevating fixed-component-shape deconstruction to a top-level type property, narrowing the narrative to deconstructible class in mail #2; JEP 468 (Candidate, preview) has long awaited a broader class-level deconstruction path. This article explains motivation, terminology alignable with public documentation, and how engineers should read preview features and the upcoming Pattern Assignment (no Preview JEP yet; see Amber features 2026 mail) in dependency order.
From RAG to search agents: BEIR co-author Nandan Thakur on BrowseComp-Plus, synthetic data pipelines, GRPO economics, and why retrieval benchmarks, training cost, and harness design pull in different directions.