A common requirement in the area of data engineering is to first process existing historical data before processing continuously live data. Processing existing data first is also referred to as bootstrapping the system. How to easily achieve this with Apache Flink? In this blog-post we will look at Flink's HybridSource which is specifically designed for such a task. If you want to clone the repository with the code from this blog post, use
Since its inception, Apache Flink has undergone significant evolution. Today, it not only serves as a unified engine for both batch and streaming data processing but also paves the way toward a new era of streaming data warehouses. Apache Flink has the concept of Dynamic Tables, which bear resemblance to materialized views in databases. However, unlike materialized views, Dynamic Tables are not directly queryable. Recognizing the need to support querying of intermediate tables
Every year, Apache Flink® sets new records in its development journey. Standing as a testament to its growing popularity, Flink now boosts over 1.6k contributors, 21k GitHub stars, and 1.4M downloads. In operational environments, Flink clusters are reaching impressive scales, with some individual clusters surpassing 2000 nodes. The largest known Flink infrastructure in production boasts over 4 million CPU cores, processing a staggering 4.1B events per second. If scalability is a concern
In October, at Flink Forward 2023, Streamhouse was officially introduced by Jing Ge, Head of Engineering at Ververica. In his keynote, Jing highlighted the need for Streamhouse, including how it sits as a layer between real-time stream processing and Lakehouse architectures, and discussed the business value it provides.
In this blog post, you will learn how to build a real-time data view on top of your Streamhouse using Apache Paimon table format. If you are coming from the Data Management world, you might know that Data engineers are generally concerned about implementing a data analytics pipeline, minimizing compute-infrastructure cost, and achieving the smallest end-to-end latency for the target users.
推荐系统是一种信息过滤系统，用于预测用户偏好，从大量的信息中筛选出用户可能感兴趣的内容进行个性化推荐。一个完整的推荐系统流程主要包括了 多路召回 -> 素材补全 -> 精排过滤 -> 混排 ->适配输出 等处理节点。混排作为结果输出前的最后一层处理，主要作用是将不同来源的推荐结果进行归一化的组合排序，一方面是为了获取对于用户推荐效果最优的排序序列，另一方面也能提高推荐的多样性、个性化以及覆盖范围。
最近经常收到内部业务方的咨询，他们想知道"如何让我们的业务系统接入大模型提效"。为了回答这个问题，我们梳理了 KubeAI 大模型平台对接的一些业务实践与一些业界经典案例分享给大家。 OpenAI 的第一次开发者大会的主题为 Maximizing LLM Performance，提出业务系统可以通过三种方式接入大模型，PROMPT（直接给大模型输入提示词），RAG（通过检索增增强来提升大模型的能力），Fine-tuning（通过微调训练来提升大模型的能力）。 本文借鉴 OpenAI 的观点，结合具体实践例子分别介绍这三种接入方式，最后建议业务可以通过渐进（PROMPT，RAG，Fine-tuning）的方式接入大模型，从而达到最佳的收益效果。
