Tempo Framework Enhancement: Zero-Downtime Deployment, AIGC Integration, and Multi-End Adaptation Practices

This article discusses the evolution and enhancement of Tempo, a CMS UI framework used by NetEase Cloud Music. As LLM applications become prevalent in business production, intelligent assistants with ChatUI interaction have become essential in middle-office scenarios.

Background

Currently, as LLM continues to be attempted and implemented in actual business production, intelligent assistants with ChatUI as the main interaction method are indispensable in middle-office scenarios. This approach of completing user requests through chat dialogue has greatly changed traditional user habits.

Since most Cloud Music CMS platforms use the Tempo framework, some platform customization and mobile office demands have emerged, posing new requirements for Tempo.

Current State

To better and faster respond to business platform customization demands, Tempo continuously abstracts platform properties to improve its configuration capabilities. Modifications to platform Logo, logo jump links, titles, footer information, layout, and custom search content can all be configured online with one click.

Since Tempo has been implemented in nearly 100+ main applications, although main applications have low release frequency, each Tempo upgrade requires main applications to update dependency versions and redeploy on Febase, which brings high upgrade costs.

Problems

Each Tempo version upgrade requires main applications to update dependency versions and redeploy - cannot achieve zero-downtime deployment

Platform functions are complex, new users have high learning costs and lack intelligent Q&A and interaction capabilities

CMS platform does not support login state integration and multi-end style adaptation , users cannot achieve mobile office

Whether for simple or complex requirements, developers need to write code - cannot achieve online configuration, plugin loading

Solutions

Zero-Downtime Deployment

Since zero-downtime deployment is the foundational capability for all new features, by summarizing and categorizing daily main application iteration requirements, the new R&D model for main applications is defined as: simple requirements online configuration, complex requirements code development and plugin loading, public features automatic upgrade.

The implementation solution involves: during Febase cloud build, external Tempo dependencies in main applications; after build completion, modify HTML content to add meta tags for storing version key parameters and real entry file paths; dynamically modify entry to loader script. This script fetches main application configuration and latest Tempo version via API, then dynamically loads Tempo UMD resource addresses and real entry file paths.

AIGC Exploration and Practice

With the emergence of ChatGPT, LLM-based logical understanding and derivation capabilities, along with conversational interaction methods, have sparked a new wave of AIGC. Compared to C-end scenarios, the complexity of interaction and business logic in middle-office scenarios makes the platform itself complex.

The Cloud Music frontend team explored and constructed a low-cost LLM service integration product solution by building basic services, converging and沉淀通用服务, UI interaction and expression methods to help businesses quickly and低成本构建知识库、智能问答、AI驱动产品功能等一系列能力.

Intelligent Recognition

Compared to NLP for user intent recognition, LLM's logical understanding and derivation capabilities are superior; it can accurately understand user input content and extract key parameters from user input using Prompt.

Intelligent Answer

After recognizing user intent, different content can be displayed based on the result type returned by the service.

Intelligent Interaction

Based on recognized user intent actions, further interaction with the platform can be performed, such as opening pages, filling form data, etc.

Multi-End Style Adaptation

To meet business personnel's mobile office demands, Tempo supports multi-end adaptive capabilities and integrates user login state with PMS. When opening CMS platform application links in Popo, direct免登成功 can be achieved.

Screen width intervals are divided into three categories: PC screen (Screen >= 1200px), Pad screen (1200px > Screen >= 768px), Phone screen (768px > Screen). Component adaptation includes Modal, Table, Form, Layout, Description.

Online Configuration

Online configuration capability is implemented based on PaaS-provided main application configuration service. Tempo further abstracts and provides default configurations for common platform properties. Configuration property priority: atomic component properties > main application online configuration > Tempo default configuration.

Plugin Loading

Asynchronous plugin loading is another capability provided by Tempo. Using main application online configuration, after configuring target plugin UMD resources, the resource script is automatically loaded during platform initialization.

Summary

Through zero-downtime deployment capability, main applications integrated with Tempo gain automatic upgrade features, directly具备多端适配和移动办公能力; by adding related configurations online, intelligent assistants can be integrated with one click, greatly reducing R&D and deployment costs caused by upgrading dependencies.