Articles

This article provides a detailed introduction to LangExtract, a Python library specifically designed for leveraging Large Language Models (LLMs) to extract structured information from unstructured text. It delves into LangExtract's core capabilities, including precise source text mapping, reliable structured output, optimized long document processing, interactive visualization, and flexible support for various LLM models across arbitrary domains. The article then demonstrates, with code examples, how to define extraction tasks, provide examples, run extraction, and visualize the results. Special emphasis is placed on utilizing LLM's world knowledge to enhance extraction quality and optimized methods for handling long documents. Furthermore, it offers installation guides (including PyPI, source code, Docker), detailed API key setup methods, and instructions on integrating OpenAI models and local Ollama models. Finally, through multiple practical application cases such as full-text extraction from 'Romeo and Juliet', drug extraction, and radiology report structuring, the article showcases LangExtract's powerful functionalities and broad applicability.

This article introduces exo, a software maintained by exo Labs, designed to integrate users' everyday devices into a powerful AI cluster. Key features of exo include automatic device discovery, immediate (day-one) support for RDMA over Thunderbolt to drastically reduce inter-device latency, topology-aware automatic parallel computing, and tensor parallelism. These features optimize model splitting and execution speed by intelligently adapting to available device resources and network conditions. The article also highlights that exo utilizes MLX as its inference backend and distributed communication framework. Benchmarking has demonstrated exo's success in running large models such such as Qwen3-235B, DeepSeek v3.1 671B, and Kimi K2 Thinking across multiple Mac Studios. The article provides comprehensive quick-start guides for running exo from source code or via a macOS application, and illustrates how to interact with the cluster through its API, covering instance location preview, model instance creation and deletion, and chat completion requests.