@subhaganapathi.bsky.social
Reposted
If you’ve built LLM workflows for structured data extraction, you’ve likely hit issues with schema drift, missing facts, and prompt brittleness. @subhaganapathi.bsky.social's new article explores how LangExtract orchestrates prompt tuning, chunking, and multi-pass validation.
Extracting Structured Data with LangExtract: A Deep Dive into LLM-Orchestrated Workflows | Towards Data Science
A guide to building modular workflows for structured intelligence
towardsdatascience.com
September 8, 2025 at 5:04 PM
If you’ve built LLM workflows for structured data extraction, you’ve likely hit issues with schema drift, missing facts, and prompt brittleness. @subhaganapathi.bsky.social's new article explores how LangExtract orchestrates prompt tuning, chunking, and multi-pass validation.
Reposted
Accuracy is often misleading. Subha Ganapathi's article dives into logical paradoxes like Simpson's, Accuracy, and Goodhart's Law, revealing how metrics can trick you in both BI and LLM contexts.
How Metrics (and LLMs) Can Trick You: A Field Guide to Paradoxes | Towards Data Science
When numbers lie — and your metrics mislead you
towardsdatascience.com
July 16, 2025 at 7:12 PM
Accuracy is often misleading. Subha Ganapathi's article dives into logical paradoxes like Simpson's, Accuracy, and Goodhart's Law, revealing how metrics can trick you in both BI and LLM contexts.