dair.ai weekly picks: DecisionRAG replaces retrievers with a hierarchical decision model
dair_ai · x · 2026-11-02
dair.ai's top papers of the week (Oct 26–Nov 1) is led by DecisionRAG from Cornell and ASAPP. Instead of a keyword/embedding retriever, a decision model navigates the corpus hierarchically (topics → documents → passages), scoring all candidates in one forward pass without generating tokens; sequential decisions grow logarithmically with corpus size. Judging candidates with shared context adds +7.7 R@1 on MultiDoc2Dial and +6.5 Hit@1 on Natural Questions. Companion figures also show other featured work: hierarchical episodic-memory agent architectures and a four-layer verification framework for coding agents.
More from coding & agent
- Together launches post-training platform aligning RL training with serving — togethercompute · 2026-11-02
- Five controls that keep LangGraph agents alive in production: 15% of chats routed to humans — LangChain · 2026-11-02
- Building a Company OS on Claude Code: skills run in 21-85s at $0.08-0.32 per task — aakashgupta · 2026-11-02
- Building a local tool to give agents access to ReMarkable notes, skipping third-party MCPs — annetgriffin · 2026-11-02
- Developer shares hard-won lessons on publishing Android apps to Chinese app stores — dotey · 2026-11-02
- AI Researcher jxnlco Solicits Real Projects Using Models for Physical Engineering and CAD — jxnlco · 2026-11-02