Onehouse announced that Lakegres now supports native vector indexes built directly on existing Hudi and Iceberg open tables, exposing similarity search through a Postgres endpoint while keeping both index and data in open formats inside the customer's own bucket. This lets agents query lakehouse-scale logs, documents and event history without moving data into a separate application database.
Vector Search
向量检索
Technology finding nearest neighbors in multi-dimensional space based on semantic similarity.
Curated Intelligence
Google Research has open-sourced Agile Deliberation, a framework that lets users define fuzzy 'gray area' concepts before an AI classifier learns them. The system splits a concept into in-scope and out-of-scope subconcepts, surfaces borderline samples for user labeling, and has Gemini rewrite the definition as a VLM prompt. Across 18 live sessions it beat zero-shot Gemini 2.5 Flash by 10.5 F1 points on average, with code released under Apache 2.0.
X has implemented two key algorithmic changes to its 'For You' feed, aiming to maximize user dwell time. The new ranking model better predicts probable time spent, while the retrieval model prioritizes posts users are likely to read completely. This shift restructures how content from non-followed accounts is filtered and coincides with a significant growth in daily active seconds to approximately 500 billion.
Amazon publishes a tutorial demonstrating how to build an agent system using Amazon Bedrock AgentCore and the Strands Agents framework. The solution orchestrates document summarization, vector search, and image generation tools powered by Claude Sonnet 4.6 to reduce financial report-to-presentation creation time to under 90 seconds.
This article proposes a 'four-gear flywheel' engineering paradigm for building self-evolving Agents, covering evaluation, memory, implementation, and control. It identifies that most teams fail due to broken data pathways between these stages, emphasizing an automated loop where execution feeds into experience accumulation and verified improvements.
Sentence Transformers officially announced its integration into the Hugging Face ecosystem. This move aims to consolidate open-source embedding tools and strengthen community collaboration, though specific technical migration details remain undisclosed.