extraction

Boris Cherny released a 7-page PDF on Graph Engineering, showing how to use four Claude prompts to replace trained ML models by building a persistent knowledge graph for agents, with steps for extraction, resolution, summarization, and querying.

The article outlines five workflows to extract intelligence from Fable 5, a frontier AI model, before it moves from subscription to pay-per-token credits, emphasizing capturing judgment and knowledge that cheaper models cannot replicate.

Discussion about whether ML teams are actually testing model security risks like extraction and poisoning in production, noting that security review for models lags behind regular software.

Hyper-Extract is an open-source framework that converts messy documents into typed knowledge structures, supporting multiple graph architectures like GraphRAG, LightRAG, and KG-Gen, with 10+ extraction engines and 80+ YAML templates for various domains.

Taste Lab is a tool that extracts a website's design DNA, allowing users to analyze and replicate design elements.

This paper proposes SafeLLM, an extraction-based approach for retrieving information from safety-critical documents, showing that line-number selection outperforms rewriting-based RAG methods in reducing hallucinations while maintaining high recall.

The author shares an open-source tool that converts web pages to design.md. It's easy to use: just send a URL and it will extract the design file.

Parse-Flow is an open-source visual workflow designer that composes document intelligence primitives (parsing, extraction, classification, splitting) into reusable pipelines, backed by LlamaIndex and a Python worker.

AI search engines extract and paraphrase content from pages, prioritizing citeable, evidence-backed material over generic summaries. Content creators should focus on writing for citation rather than just clicks.

kg-gen is an open-source Python package that uses language models (via LiteLLM and DSPy) to extract knowledge graphs from plain text or conversation messages, featuring chunking, clustering, and flexible provider routing.

Someone created a magic tool designmd.me that lets you extract a complete design system (colors, fonts, components, layout) by simply entering a website URL, automatically generating a DESIGN.md file. Combined with the awesome-design-md repository (already 54k stars), you can quickly replicate the UI of top brands.