UMass AutoIndex Turns Chunking Into Programs an LLM Writes, Not Configs You Tune
mrdrozdov · x · 2026-09-14
Weaviate Podcast #143 features Sam O'Nuallain (UMass Amherst) on AutoIndex, which treats indexing as code optimization rather than a config you tune.
- An analysis agent and a code agent loop together to write Python "representation programs" that chunk, enrich, and reorganize your corpus.
- Every hypothesis must prove validation lift before it survives.
- Key lesson: "did recall go up?" is useless feedback — giving the analysis agent tools to investigate why a gold document ranked low is what made the system work.
A fresh paradigm for RAG practitioners.
Related event: AutoIndex Optimizes Retrieval with LLM-Written Indexing Code(2 posts)→
More from coding & agent
- Astra-built racer spawns a ghost of your last lap every time you finish one — LexicalLegend · 2026-09-14
- Alchemy Console ships a visual UI for browsing and deleting Alchemy cloud resources — samgoodwin89 · 2026-09-14
- Edit video for free: pair Codex Astra 6 with the free DaVinci Resolve 19.1 — alexcovo_eth · 2026-09-14
- Multi-agent RAG lesson: structured JSON schemas plus runtime verification to stop hallucinated API calls — kashifmanzoor · 2026-09-14
- A security layer for agent infra: replay attacks in sandboxes, learn from them — wandb · 2026-09-14
- Muse Spark 1.3 wins fans: browser agent files insurance claims faster than humans — alexandr_wang · 2026-09-14