AutoIndex: LLM agents write Python programs to replace tuned chunking for search
CShorten30 · x · 2026-09-14
The 143rd Weaviate Podcast features Sam O'Nuallain (UMass Amherst) on AutoIndex, a research project that rethinks how data is represented for search:
- Core idea: chunking strategy isn't a config you tune but a program an LLM writes. An analysis agent and a code agent loop together to produce Python "representation programs" that chunk, enrich, and reorganize your corpus
- Admission rule: every hypothesis must prove validation lift before it survives—indexing is treated as code optimization
- Biggest 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
Rather than heuristic approaches, AutoIndex learns retrieval representations directly as a task—an intriguing convergence of IR and agents.
Related event: AutoIndex: LLM-Written Code Optimizes Retrieval Indexing(3 posts)→
More from coding & agent
- Building games with AI? Set up a 'tech art' test scene to tune your look fast — chongdashu · 2026-10-31
- Skydive opens to all: watch your AI coworkers work in a 3D office — Scobleizer · 2026-10-31
- Handwritten RAGEN code becomes the #3 hardest task in FrontierCS 2 — wzenus · 2026-10-31
- LangChain SF meetup: decision models and a third path for agent evals — LangChain · 2026-10-31
- Cap your coding agent's retries at 5: how SuperPlane stops runaway token bills — Arindam_1729 · 2026-10-31
- Classical infrastructure gets more important, not less, in the age of AI agents — graceisford · 2026-10-31