2025-11-18 Tech Briefing | Frontend Creative Drawing Practice

Tuesday

Welcome to this edition, featuring 23 articles.

📚 Contents/📝 TL;DR

📚 Contents

📝 TL;DR

Primary

🧸 Frontend is Not Just for Writing Management Backends, I Drew a LABUBU with 400 Lines of Code!

Frontend Creative Drawing Practice

🧸 Frontend is Not Just for Writing Management Backends, I Drew a LABUBU with 400 Lines of Code!

Frontend Creative Drawing Practice

WeChat Mini Game Package Size Limit of 4M, One Font Alone Takes 11.24M, How to Play?

Font Minification Technique Analysis

WeChat Mini Game Package Size Limit of 4M, One Font Alone Takes 11.24M, How to Play?

Font Minification Technique Analysis

openEuler Releases a Super Node Operating System, Set to Launch by Year-End! Xiong Wei: The Future of Operating Systems Points to Natural Language in the AI Era

openEuler releases an operating system for super nodes, set to launch by year-end

openEuler Releases a Super Node Operating System, Set to Launch by Year-End! Xiong Wei: The Future of Operating Systems Points to Natural Language in the AI Era

openEuler releases an operating system for super nodes, set to launch by year-end

Efficient Comparison of Two Amazon DynamoDB Tables

Introduces an efficient algorithm and tool for comparing data differences between two DynamoDB tables.

Efficient Comparison of Two Amazon DynamoDB Tables

Introduces an efficient algorithm and tool for comparing data differences between two DynamoDB tables.

AI Gateway Proxy for RAG Retrieval: Dify's New Practice for Integrating External Knowledge Bases

Dify's new method for integrating with external RAG engines

AI Gateway Proxy for RAG Retrieval: Dify's New Practice for Integrating External Knowledge Bases

Dify's new method for integrating with external RAG engines

🔥 The Most Ironical Job in the Large Model Era: 'Large Model Aftercare Engineer'

AI handles 80% but the last 20% requires human aftercare, creating a new role: 'Aftercare Engineer'.

🔥 The Most Ironical Job in the Large Model Era: 'Large Model Aftercare Engineer'

AI handles 80% but the last 20% requires human aftercare, creating a new role: 'Aftercare Engineer'.

Anthropic AI Spy Activity Exposure: Distinguishing Signals from Noise

The core issue of the Anthropic exposure is the risk of AI tool jailbreaking.

Anthropic AI Spy Activity Exposure: Distinguishing Signals from Noise

The core issue of the Anthropic exposure is the risk of AI tool jailbreaking.

From Core Technologies to Full-Process Implementation: HarmonyOS Native Development and AI Efficiency Practices | HarmonyOS Leader's Talk

Sharing of HarmonyOS native development technology and AI efficiency practices

From Core Technologies to Full-Process Implementation: HarmonyOS Native Development and AI Efficiency Practices | HarmonyOS Leader's Talk

Sharing of HarmonyOS native development technology and AI efficiency practices

Essential

Replicate Joins Cloudflare

Replicate Acquired by Cloudflare, Integrated into Workers AI

Replicate Joins Cloudflare

Replicate Acquired by Cloudflare, Integrated into Workers AI

CSS Also Supports if Now!!! CSS if() Function is Here!

CSS introduces the if() function, enabling conditional judgments

CSS Also Supports if Now!!! CSS if() Function is Here!

CSS introduces the if() function, enabling conditional judgments

Frontend Graphics Engine Architecture Design: A Scalable Dual Rendering Engine Supporting Custom Renderers

Architecture design of a dual rendering frontend graphics engine based on ECS.

Frontend Graphics Engine Architecture Design: A Scalable Dual Rendering Engine Supporting Custom Renderers

Architecture design of a dual rendering frontend graphics engine based on ECS.

🔥 The Most Ironical Profession in the Era of Large Models: 'Large Model Aftercare Engineers'

AI can handle 80% but the final 20% requires human aftercare, giving rise to 'aftercare engineers'.

🔥 The Most Ironical Profession in the Era of Large Models: 'Large Model Aftercare Engineers'

AI can handle 80% but the final 20% requires human aftercare, giving rise to 'aftercare engineers'.

SLS Copilot Practice: Building a Data Infrastructure for LLM Applications with SLS

Leverage Alibaba Cloud SLS to build an observable data infrastructure for LLM applications, addressing the limitations of the Dify platform.

SLS Copilot Practice: Building a Data Infrastructure for LLM Applications with SLS

Leverage Alibaba Cloud SLS to build an observable data infrastructure for LLM applications, addressing the limitations of the Dify platform.

Config-Driven Dynamic Agent Architecture Networks: Achieving Efficient Orchestration, Dynamic Updates, and Intelligent Governance

Deep dive into the agent architecture design paradigm

Config-Driven Dynamic Agent Architecture Networks: Achieving Efficient Orchestration, Dynamic Updates, and Intelligent Governance

Deep dive into the agent architecture design paradigm

Using AI to Accelerate Database Modernization: Database Modernizer Workshop

AWS introduces an AI-driven database migration workshop

Using AI to Accelerate Database Modernization: Database Modernizer Workshop

AWS introduces an AI-driven database migration workshop

Frontend Real-time Push & WebSocket Interview Questions (2026 Edition)

Comprehensive analysis of frontend real-time push technologies and WebSocket architecture along with key interview points.

Frontend Real-time Push & WebSocket Interview Questions (2026 Edition)

Comprehensive analysis of frontend real-time push technologies and WebSocket architecture along with key interview points.

A Deep Dive into Android Crash Capture Principles and Closed-Loop Practices from Crash to Attribution

Detailed explanation of Android crash capture and closed-loop attribution implementation.

A Deep Dive into Android Crash Capture Principles and Closed-Loop Practices from Crash to Attribution

Detailed explanation of Android crash capture and closed-loop attribution implementation.

How to Use Domain-Driven Design and Change Stalking for Socio-Technical Design

Domain-Driven Design combined with Socio-Technical Design provides a method to drive organizational change through dialogue and small experiments.

How to Use Domain-Driven Design and Change Stalking for Socio-Technical Design

Domain-Driven Design combined with Socio-Technical Design provides a method to drive organizational change through dialogue and small experiments.

General

Why Are Color Websites Charging? How Much Would It Cost to Build One?

Detailed breakdown of the cost components for building a video website.

Why Are Color Websites Charging? How Much Would It Cost to Build One?

Detailed breakdown of the cost components for building a video website.

CloudSpeedBuild AI Assistant Launched: Conversational Generation of Deployable Alibaba Cloud Architecture Diagrams

AI assistant simplifies cloud architecture design

CloudSpeedBuild AI Assistant Launched: Conversational Generation of Deployable Alibaba Cloud Architecture Diagrams

AI assistant simplifies cloud architecture design

Automating AWS VPN Maintenance with Tunnel Endpoint Lifecycle

AWS releases automated VPN tunnel maintenance solution

Automating AWS VPN Maintenance with Tunnel Endpoint Lifecycle

AWS releases automated VPN tunnel maintenance solution

HarmonyOS DevEco Studio Core Technologies! Application Development Efficiency Maximized

DevEco Studio enhances HarmonyOS development efficiency through multi-platform debugging and intelligent code completion.

HarmonyOS DevEco Studio Core Technologies! Application Development Efficiency Maximized

DevEco Studio enhances HarmonyOS development efficiency through multi-platform debugging and intelligent code completion.

No Need to Hand-Craft Code Entirely! CodeGenie Accelerates the Realization of Ideas!

HarmonyOS AI-assisted programming tool CodeGenie is released.

No Need to Hand-Craft Code Entirely! CodeGenie Accelerates the Realization of Ideas!

HarmonyOS AI-assisted programming tool CodeGenie is released.

Primary

🧸 Frontend is Not Just for Writing Management Backends, I Drew a LABUBU with 400 Lines of Code!

Juejin Frontend Hottest•Published on November 13

Insight•Frontend Development•Score: 9/10

LABUBUCanvas DrawingLeaferJSFrontend Creativity

Source

◆

WeChat Mini Game Package Size Limit of 4M, One Font Alone Takes 11.24M, How to Play?

Juejin Frontend Hottest•Published on November 14

Insight•Frontend Development•Score: 8/10

WeChat Mini GameFont MinificationFontminPackage Size Optimization

Source

◆

openEuler Releases a Super Node Operating System, Set to Launch by Year-End! Xiong Wei: The Future of Operating Systems Points to Natural Language in the AI Era

InfoQ - Backend•Published on November 17

News•Cloud Computing and Backend•Score: 8/10

openEulerSuper NodeAI Operating SystemNatural Language InteractionLingqu Interconnection

Source

◆

Efficient Comparison of Two Amazon DynamoDB Tables

AWS - Databases•Published on November 17

Insight•Cloud Computing & Backend•Score: 8/10

DynamoDBData SynchronizationData ComparisonAWS GlueNoSQL

🔬

Tech Insights

The core highlight is that it solves a major challenge in data validation/consistency checks for distributed NoSQL databases. DynamoDB scans are sorted by keyspace, making 'diff' operations feasible. By utilizing internal DynamoDB features like partition key hash consistency and the predictability of parallel scans, the approach transforms large-scale table comparisons into independent and parallel tasks, achieving linear time complexity and minimal memory usage. This is a textbook example of 'breaking the problem with cleverness', worth learning for all engineers handling massive data. The article's concept of 'scan stream' abstracts pagination scans into continuous streams, significantly simplifying the algorithm logic. This mindset is valuable for any data comparison involving paginated APIs. A practical case showed two tables with 500 million entries being compared in 6.5 minutes for less than $10, which is remarkable in terms of performance and cost efficiency. This is a blessing for scenarios requiring frequent data validation, such as post-blue-green deployment checks.

⚠️

Notable

The advantage of this approach is built on DynamoDB's underlying features, which limits its reference value for non-DynamoDB users or those using other NoSQL databases (e.g., Cassandra, MongoDB). Although the algorithm is clever, the actual use case may result in massive diff outputs if the table schemas are inconsistent or the data volume and differences are large, and the article does not elaborate much on how to effectively handle and present these differences. In practice, difference handling is often more complex than difference detection. Regarding RU (read/write units) costs for DynamoDB itself, our team found that if not using On-Demand mode but Provisioned mode with improper configuration, it's easy to waste resources or hit performance bottlenecks. While the article mentions this, it could go deeper into optimizing RU usage during comparison operations, such as whether Reserved Capacity could be leveraged.

Source

◆

AI Gateway Proxy for RAG Retrieval: Dify's New Practice for Integrating External Knowledge Bases

Alibaba Cloud Tech•Published on September 3

Insight•AI and Frontier Technologies•Score: 8/10

RAG retrievalDifyHigress gatewayknowledge base integrationAI enhancement

Source

◆

🔥 The Most Ironical Job in the Large Model Era: 'Large Model Aftercare Engineer'

Juejin Frontend Hottest•Published on November 17

Insight•AI and Frontier Technologies•Score: 8/10

Large ModelAftercare EngineerAI AgentSoftware EngineeringHuman-AI Collaboration

🔬

Tech Insights

The article's description of the '80% illusion' of large models is spot-on. It 'cuts to the chase' by pointing out the critical flaws in AI's lack of product logic, business context, and security awareness. This mirrors the issues our team recently faced when trying to use AI for development assistance. The article classifies AI Agents into two categories: 'Workflow Type' and 'Autonomous Type', and clearly states that the former is 'practical and deployable' while the latter is 'free-spirited, but a disaster in reality'. This classification is very useful and can guide our selection of appropriate AI use cases. The role of 'Aftercare Engineer' is well-defined, covering proofreading, refactoring, and polishing. These three essential skills are the key to transforming AI outputs into 'production-grade products'. It emphasizes the value of experience, thinking, and product understanding, offering meaningful guidance for team members' career paths.

Source

◆

Anthropic AI Spy Activity Exposure: Distinguishing Signals from Noise

Thoughtworks Tech•Published on November 18

Insight•AI Security and Defense•Score: 7/10

AI JailbreakingAI SecurityAttack SurfaceModel AlignmentSecurity Automation

Source

◆

From Core Technologies to Full-Process Implementation: HarmonyOS Native Development and AI Efficiency Practices | HarmonyOS Leader's Talk

InfoQ - Frontend•Published on November 11

Insight•Frontend Development•Score: 7/10

HarmonyOS Native DevelopmentArkTSDevEco StudioAI-assisted ProgrammingCross-device Collaboration

Source

Essential

Replicate Joins Cloudflare

Cloudflare Blog•Published on November 17

News•AI and Frontier Technologies•Score: 10/10

ReplicateCloudflareAI Inference PlatformModel as a ServiceEdge AI

Source

◆

CSS Also Supports if Now!!! CSS if() Function is Here!

Juejin Frontend Hottest•Published on November 12

Insight•Frontend Development•Score: 9/10

CSS if()conditional logicstyle queriesmedia queries

Source

◆

Frontend Graphics Engine Architecture Design: A Scalable Dual Rendering Engine Supporting Custom Renderers

Juejin Frontend Hottest•Published on November 14

Insight•Software Architecture and Engineering Practices•Score: 9/10

ECS ArchitectureDual Rendering EngineCanvasKitFrontend Graphics EnginePluginized Design

Source

◆

🔥 The Most Ironical Profession in the Era of Large Models: 'Large Model Aftercare Engineers'

Juejin Frontend Hottest•Published on November 17

Insight•AI and Frontier Technologies•Score: 9/10

Large ModelsAftercare EngineerAI AgentSoftware EngineeringHuman-Machine Collaboration

🔬

Tech Insights

The article's description of the '80-point illusion' of large models is very accurate, pointing directly to critical flaws such as the lack of product logic, business context, and safety awareness. This mirrors the issues our team recently encountered while internally testing AI-assisted development. The article categorizes AI agents into two types: 'workflow-based' and 'autonomous', emphasizing that the former is practical and deployable, while the latter is 'free-spirited, real disaster'. This classification is highly useful and can guide the selection of appropriate AI application scenarios. The role of 'aftercare engineers' is clearly defined, with the three essential tools—proofreading, restructuring, and refining—being the key to transforming AI outputs into production-grade products. It underscores the value of experience, thinking, and product understanding, offering valuable guidance for team members' career development.

Source

◆

SLS Copilot Practice: Building a Data Infrastructure for LLM Applications with SLS

Alibaba Cloud Tech•Published on September 28

Insight•AI and Frontier Technologies•Score: 9/10

LLM applicationsobservabilitySLSDifydata infrastructure

📝

Summary

This article discusses how the Alibaba Cloud SLS team utilized Alibaba Cloud SLS to resolve three key pain points in LLM application observability: insufficient query capabilities, missing traceability, and unhelpful content presentation. By implementing dual writing of data to SLS and reengineering the data infrastructure, they achieved capabilities such as elastic scalability, multi-dimensional querying, and real-time analysis. Combined with OneDay and DingTalk AI Table, they built a lightweight diagnostic system and a quality optimization loop. For our team, this highlights that in the current LLM application development boom, data infrastructure and observability are the foundation for ensuring application quality and iteration efficiency, and timely follow-up and planning for LLM observability are crucial.

🔬

Tech Insights

The article directly addresses the core pain points of LLM application development: while model tuning and feature implementation are important, the lack of observability is the real bottleneck. Using the Dify case, it demonstrates the scalability, query performance, and real-time analysis limitations of traditional OLTP databases (PostgreSQL) when handling massive, complex, and dynamically changing log data. This mirrors our team's challenges in traditional application monitoring, but is even more pronounced in LLM scenarios. Alibaba Cloud SLS, as a cloud-native log service, offers advantages such as elasticity, pay-as-you-go, full-text search, multi-dimensional querying, and streaming processing, which perfectly align with the data characteristics of LLM applications. The gradual architecture evolution strategy from PostgreSQL dual writing to SLS is very stable and worth learning. It also emphasizes SLS's native support for loose schemas and diverse data formats (JSON/Markdown), which are like a 'lifeline' for LLM applications with multimodal and structurally unpredictable outputs. The article mentions the 'idea + data + AI' new development model, and through the integration of OneDay (Alibaba's internal AICoding platform) and DingTalk AI Table, it achieves a minimalist architecture, building a lightweight tracing system in one day. This clearly indicates the trend of toolchain integration and data-driven quality loops in the AI era—using SLS to collect data, AI tables for manual annotation and analysis, and then feeding back to the LLM model for optimization, forming a closed loop. This is something all AI product managers and architects should consider.

⚠️

Notable

The article emphasizes the power of SLS, but it lacks deeper details on how to extract and standardize log data from Dify or other LLM frameworks, and how to design and optimize LogStore in SLS to support efficient querying and analysis. In practice, this is often a challenging aspect of data ingestion and governance. OneDay and DingTalk AI Table are internal tools at Alibaba, although they showcase the potential of the 'idea + data + AI' approach, they pose significant implementation challenges for external teams, requiring the search for alternative general-purpose AI tools or self-research. This shortcoming makes the promotion of universal experience less effective. The article mentions 'content security' as a challenge for LLM applications, but it focuses only on traditional network attacks. It does not explore in depth how to monitor and protect against Prompt Injection, model hallucination, and data leaks, which are specific security concerns for LLMs. This is a critical consideration for LLM applications today.

📈

Perspectives

By 2025, LLM applications have entered deeper waters, and pure model capabilities are no longer the sole barrier. Engineering and stability are the key factors determining whether a product can be successfully deployed. This Alibaba Cloud practice precisely hits the industry pain points. It sends a clear signal: cloud providers are actively offering full-stack support for LLM applications, from models (Baichuan) to infrastructure (SLS). For other cloud providers and SaaS platforms, 'LLM observability as a service' will represent a significant market opportunity. In the future, the maturity of an LLM product will no longer be judged solely by its content generation quality, but also by whether it has a robust data infrastructure that supports rapid iteration, problem diagnosis, and continuous optimization. Teams still relying on manual annotation and debugging will soon be outpaced by the market.

Source

◆

Config-Driven Dynamic Agent Architecture Networks: Achieving Efficient Orchestration, Dynamic Updates, and Intelligent Governance

Alibaba Cloud Tech•Published on September 23

Insight•AI and Frontier Technologies•Score: 9/10

Agent ArchitectureConfiguration-DrivenDynamic UpdatesA2A ProtocolRAG Knowledge Base

Source

◆

Using AI to Accelerate Database Modernization: Database Modernizer Workshop

AWS - Databases•Published on November 17

Insight•AI and Frontier Technologies•Score: 9/10

Database ModernizationAI MigrationDynamoDBMCP ServerZero Downtime Migration

Source

◆

Frontend Real-time Push & WebSocket Interview Questions (2026 Edition)

Juejin Frontend Hottest•Published on November 16

Insight•Software Architecture and Engineering Practices•Score: 9/10

WebSocketReal-time PushSSESystem ArchitectureInterview

🔬

Tech Insights

This content is truly the 'Grail' of real-time communication, with many highlights: 1. Clear technical evolution path: From full page refresh to short polling, long polling, Streaming, SSE, and finally WebSocket, it clearly demonstrates the development trajectory of real-time communication technologies, helping readers understand the background and problem-solving capabilities of each technology. This is crucial for systematic learning. 2. Horizontal scalability system design: It proposes a three-layer architecture pattern of 'gateway layer - pipeline layer - storage layer', detailing the responsibilities, selection (Kafka/Pulsar/NATS/Redis Streams), sharding strategies, and routing methods of each layer. This is extremely valuable for building high-concurrency real-time systems, directly offering production-level solutions. 3. Interview questions with practical value: The article organizes content in the form of '2026 Edition Interview Questions', covering almost all high-frequency topics from technical comparison, system design, to fault handling, performance evaluation, and authentication. If a candidate can explain these questions clearly, they are basically the 'architect candidate' I'm looking for.

⚠️

Notable

If we look for nitpicks: 1. Although the content is deep and broad, some technical details, such as the underlying optimizations of HTTP/2 and HTTP/3 on WebSocket, could be slightly more detailed. For example, how HTTP/2's multiplexing reduces head-of-line blocking for WS, or how WebTransport's advantages based on QUIC can be truly demonstrated in specific scenarios like cloud gaming and real-time audio/video. 2. Despite being in the form of interview questions, for some complex concepts like the specific use cases and trade-offs of different message buses in the 'pipeline layer', adding some industry case studies would provide readers with more intuitive understanding. For instance, Kafka's advantages in log and event stream scenarios, Redis Streams' strengths in lightweight queue use cases, etc.

Source

◆

A Deep Dive into Android Crash Capture Principles and Closed-Loop Practices from Crash to Attribution

Juejin Frontend Hottest•Published on November 17

Insight•Software Architecture & Engineering Practice•Score: 9/10

Android Crash CaptureNative CrashMinidumpSymbolizationUncaughtExceptionHandler

Source

◆

How to Use Domain-Driven Design and Change Stalking for Socio-Technical Design

InfoQ - Architecture•Published on November 18

Insight•Software Architecture and Design•Score: 8/10

Domain-Driven DesignSocio-Technical DesignChange StalkingSoftware ArchitectureOrganizational Complexity

Source

General

Why Are Color Websites Charging? How Much Would It Cost to Build One?

Juejin Frontend Hottest•Published on November 17

Insight•Software Architecture and Engineering Practice•Score: 7/10

Video WebsiteCost EstimationCloud ServicesArchitecture DesignTraffic

🔬

Tech Insights

The article strips down the cost structure of a video website to its core, covering everything from servers to databases, CDN, transcoding, and object storage. The highlight is that it clearly states traffic costs and human resource costs are the major expenses, especially for video, which can easily bankrupt a company with just casual browsing. This differs significantly from our usual approach to application service cost structures, offering direct reference value for evaluating the feasibility of video-related projects. It emphasizes content security (illegal videos) and network attacks (DDoS, SQL injection) as the two core security threats for video websites. The author provides step-by-step recommendations for upgrading from free to commercial-level protection, which is very practical. For teams in the consumer-facing business, security is always a red line, and this article offers a lot of valuable insights. The article mentions 'video on demand services' as a turnkey solution, which can significantly reduce development complexity, but the cost is merely a consolidation of various fees without any fundamental reduction. This aligns with my understanding of SaaS products, which lower the threshold but do not reduce overall operational costs.

Source

◆

CloudSpeedBuild AI Assistant Launched: Conversational Generation of Deployable Alibaba Cloud Architecture Diagrams

Alibaba Cloud Tech•Published on August 25

Insight•Cloud Computing and Backend•Score: 7/10

Cloud Architecture DesignAI AssistantCADTNatural Language InteractionCloud Migration Efficiency

Source

◆

Automating AWS VPN Maintenance with Tunnel Endpoint Lifecycle

AWS 网络•Published on November 17

Insight•Cloud Computing and Backend•Score: 6/10

AWS VPNAutomated OperationsEventBridgeLambdaCloudFormation

Source

◆

HarmonyOS DevEco Studio Core Technologies! Application Development Efficiency Maximized

InfoQ - Frontend•Published on November 13

News•Comprehensive News•Score: 1/10

HarmonyOSDevEco StudioDevelopment Efficiency

Source

◆

No Need to Hand-Craft Code Entirely! CodeGenie Accelerates the Realization of Ideas!

InfoQ - Frontend•Published on November 13

News•Comprehensive News•Score: 1/10

HarmonyOSAI-assisted programmingCodeGenie