Loading / 加载中
Thinkgap feed
Curated items are read from the processed items table and served as a bilingual feed.
6 items
Pinecone and Pulumi are co-hosting an evening of talks on June 18 at 5 PM in NYC. The event will cover the infrastructure behind vector search and retrieval-augmented generation (RAG), infrastructure as code (IaC) practices, and a demo of an AI running coach Slack bot that incorporates real-world data into model context. The program includes demos, a Q&A session, and a hangout.
Unlike most AI agents that reset every session, Jenova AI agents persist user context, with the longest session spanning 16 million tokens. All session data is retrievable in under 10 milliseconds via Pinecone vector retrieval. This persistent knowledge layer enabled the company to reach over $1M in annual recurring revenue, 200,000+ users, and 10x revenue growth in 5 months, almost entirely through organic growth. Founder Boris Wang stated that Pinecone's knowledge layer is the foundation determining user retention, calling it the product's moat.
Pinecone shares an n8n template for building a RAG pipeline using Apify, Pinecone, and Gemini. The pipeline automatically scrapes website content, indexes it into Pinecone as vector embeddings, and retrieves relevant context for answer generation. This enables a support chatbot that stays current without manual data wrangling. The template is linked in the post along with a deeper blog post from Apify.
Over four weeks, three enterprise customers exhibited the same pattern: most inference spending goes to retrieval loops. A generic index lacks domain-specific knowledge, query types, and task structure, causing the retrieval loop to run before the model can reason. Pinecone's Nexus compiles knowledge before the query to address this inefficiency. The full results are available at the provided link.
Pinecone announced the integration of its knowledge engine Pinecone Nexus with Microsoft OneLake at MSBuild. This integration allows users to build a reliable, production-grade knowledge layer over structured data in OneLake or Fabric. Staff Data Engineer Simon Lu demonstrates the process in a quick demo. The full announcement and demo are available in the provided link.
LlamaIndex announces its PDF parsing capability with a high-profile billboard campaign across San Francisco. The company invested a significant amount (seven figures) in marketing. This highlights their focus on document parsing for large language model applications. It's a promotional tweet from their official account.