Steering Semantic Data Processing With DocWrangler
summary
DocWrangler is an IDE that layers on DocETL to turn LLM calls into repeatable map/reduce-style pipelines, letting users inspect outputs, annotate them, and automatically turn those annotations into better operations. It exposes In-Situ User Notes to capture long‑tail errors, uses LLM-Assisted Prompt Refinement to convert notes into improved prompts, and runs LLM-Assisted Operation Decomposition to split tasks an LLM can’t generalize. Evaluation: 10-person study plus 1,500+ public sessions show users creating classifier-like prompts and deliberate exploration strategies.
AI-written summary · about
related talks
32mDocuments Meet LLMs: Tales from the Trenches
37mAdrian Brudaru | Data Track | Compass AI & Data Summit 2025
38mThe Art of Compression: Crafting Insightful Summaries with LLMs
33mDesigning LLM Applications for Massive Datasets
15mFrom Words to Data: Using LLMs to Turn Questions into SQL
47mValidating LLMs – A True Story
49mNext-Gen Documentation with AI- 49mNext-Gen Documentation with AI