Design and Evolution of Bilibili's Activity Platform Low‑Code System

This article presents the full lifecycle of the new Activity Platform built for Bilibili, a low‑code platform that serves both content creators (operations), end users, and developers. It describes the platform’s purpose, service objects, and three business modes: No Code (pure configuration), Low Code (semi‑custom code blocks), and Pro Code (full custom development).

The platform underwent a two‑year redesign and iterative reconstruction, resulting in a comprehensive upgrade across four layers: interface interaction, business functionality, system architecture, and development ecosystem. Key architectural components include the Eva front‑end service, Eva‑server node service, page‑service routing, and the component ecosystem.

Major challenges addressed include:

Complex UI/UX issues in the canvas and component list.

Inconsistent rendering between B‑side (editor) and C‑side (preview) environments.

Highly coupled modules that hindered independent testing and extension.

Scalability constraints of the existing component and data‑source ecosystems.

To solve these, the team introduced:

A modular, decoupled architecture with clear service boundaries.

An iframe‑based canvas with a previewWarp overlay to enable cross‑layer drag‑and‑drop.

A unified renderer (eva‑render) supporting React, Vue, and vanilla JS, ensuring identical behavior on both editor and runtime.

Re‑engineered layer tree using react‑DnD for more precise and extensible drag interactions.

A data‑source panel that consolidates configuration, reducing errors and improving efficiency.

The development was organized into three phases:

Phase 1 – Short‑term optimization: UI refresh, canvas improvements, and data‑source integration, achieving a 20%+ efficiency gain for operators.

Phase 2 – Long‑term architectural upgrade: Service refactoring, renderer unification, component conversion tools, and the introduction of eva‑cli for No Code, Low Code, and Pro Code workflows.

Phase 3 – Finalization and performance tuning: Server‑side pre‑rendering, runtime optimizations, and extensive metrics collection, resulting in a 35% reduction in first‑contentful‑paint for large events.

Quantitative outcomes include:

Average activity build time reduced from 3 person‑days to 0.48 person‑days per activity (≈20 hours).

Platform covered 77% of Bilibili’s activities by H1 2024, delivering over 8.4 billion page views.

Key events such as “Black Myth: Wukong” achieved >10 million PV and >180 k submissions.

Future directions focus on intelligent quality inspection, automated testing, integration of multimodal large language models for design‑to‑code, and data‑driven activity effectiveness evaluation.