Tag
NVIDIA 发布预训练关系型基础模型 Kumo Relational，可直接在企业数据库上通过单次前向传播完成预测，无需训练与特征工程；在 41 项任务与 15 个数据库上超越监督式关系深度学习及专家特征的梯度提升树管线。
LLM-based agents work better with structured data, as demonstrated by Harvey's context graphs for long-horizon legal tasks on the Legal Agent Bench; the author notes DAIR.AI is exploring similar approaches for research tasks.
推荐使用 Marketing Skills——一套开源的 AI 代理营销技能集合，涵盖 SEO、结构化数据优化、转化率优化等任务，可配合 Claude Code、Codex、Cursor 等编码代理使用。
DataWeave is a system combining conversational interaction, schema grounding, analytical planning, and executable query generation to support human–LLM exploratory analysis of structured datasets, validated in a case study with journalists analyzing the U.S. Department of Education's IPEDS data.
Datalab released 'lift', an open-weights 9B vision model that extracts schema-constrained JSON directly from PDFs and images in a single step, matching Gemini Flash 3.5 accuracy at 3x the speed. It handles multi-page documents, is self-hostable via vLLM or Hugging Face, and is available under Apache-2.0 code / OpenRAIL-M model licenses.
LlamaIndex 发布 Extract v2.5，推出新一代基于 schema 的文档抽取 agent，可跨多页表格进行结构化推理并提供高级引用定位，在 ExtractBench 各档位上准确率与 grounding 分数均大幅提升，且每页定价不变。
Formalini 是一款新产品，可将任意文档、照片、扫描件或手写笔记快速转换为 JSON、Excel、表格等结构化数据，并支持个人文档管理、企业工作流集成（UiPath、SAP、Salesforce）以及供 AI 代理调用的 MCP 服务器。
The article highlights the inefficiency and rationalization issues with using LLMs as judges, introduces Jev as a tool that leverages structured data for more reliable results, and questions the potential for open-source alternatives.
A solo founder reduced client complaints on automated reports by moving prose writing from an AI agent to a dedicated content generator, focusing the agent on reasoning and structured findings.
TypeSafe AI has launched Jev, a structured alternative to LLMs for automation that offers significantly lower latency and cost, with a $40 million seed round from DCVC.
LimiX-2 is a pretrained foundation model for structured data that uses contextual mechanism networks to achieve #1 on major tabular benchmarks, supporting multiple tasks without task-specific parameter updates.
A LinkedIn paper reveals that agent memory is not automatically portable across model swaps, with fixed-schema memory being more stable than free-form notes, and emphasizes the need for memory compatibility tests during upgrades.
LlamaIndex has launched Extract Turbo, a feature that enables lightning-fast extraction of structured data from documents with high speed and accuracy.
TEFM is a framework that reduces token usage for LLMs processing structured data by compressing observations into compact Behavioral Code tokens, while ensuring faithful predictions through a dual-fidelity optimization approach.
This paper introduces persistent discovery context, a lightweight memory layer for data-centric agents that stores prior intent-to-object mappings to enhance retrieval quality across tasks, demonstrating improvements in structured data environments.
This tutorial from freeCodeCamp explains how to design schemas, validate outputs, and handle failures to reliably extract structured data from LLMs, covering techniques like constrained outputs, retry loops, and streaming.
LlamaParse has introduced native agentic spreadsheet extraction using a tuned model and harness for schema-guided extraction, enabling conversion of dense sheets like balance sheets into clean structured fields.
HyperExtract is an LLM-powered framework that converts unstructured documents into structured knowledge graphs, hypergraphs, and lists, simplifying knowledge extraction and management.
This paper introduces a structured approach to extracting persona information for LLM-based digital twins, showing that structured representations improve predictive accuracy over raw transcripts. An automatic pipeline is proposed to adapt structures to different tasks.
LLMagnet is a product that enhances WordPress site visibility to AI agents by tracking AI crawler visits, generating AI-friendly data like llms.txt, and improving AI accessibility.