Key Takeaways from JD & Baidu Frontend Salon: Cross‑Platform, AI‑Native, KMM

Event Overview

On July 18, 2023, the JD & Baidu Frontend Technology Salon (Technology Fusion·Frontend Unlimited) concluded in Beijing. Organized by JD Retail Technology Committee’s frontend channel and featuring Baidu experts, the event focused on large language model (LLM) applications, edge intelligence, low‑code, and cross‑platform technologies, discussing how developers can leverage new tools to drive business growth.

JD’s Self‑Developed Cross‑Platform Engine

Speakers compared browser, native, and Flutter‑style rendering, highlighting the advantages of each. JD presented its proprietary cross‑platform engine, detailing its architectural and workflow design, which received strong approval from attendees who look forward to its real‑world deployment.

Baidi TalosLite Lightweight Cross‑Platform NA Dynamic Solution

Liao Huanyu introduced TalosLite, a framework that eliminates first‑screen jitter, achieves fast rendering, rapid rollout, and simple interaction by removing the JavaScript engine. It uses JSON as a DSL, embeds basic UI components and atomic events, and leverages an expression engine with server‑side pre‑warming to boost client‑side speed. TalosLite is widely used in Baidu’s anti‑epidemic, lifestyle mall, and digital‑human projects.

Baidi KMM Exploration in Mobile App Development

Yuan Hangguang outlined the challenges of maintaining synchronized business logic across iOS and Android, such as high coding costs and testing overhead. The team evaluated Kotlin Multiplatform Mobile (KMM) against other cross‑platform frameworks, emphasizing its smaller package size, low entry barrier, and strong inter‑operability. He demonstrated KMM integration steps, API usage, multithreading models, state sharing, atomic classes, and core libraries for networking and databases. After extensive trials in Baidu Matrix App, KMM now powers over 100,000 lines of code in Baidu’s main app.

JD’s ChatGPT‑Powered Frontend Efficiency Practices

JD explained the meaning of GPT (Generative Pre‑trained Transformer) and shared how they harness ChatGPT to improve coding experience. Their internally built VSCode plugin, HiBox, streams data‑driven suggestions, knowledge‑base Q&A, and immersive coding assistance, achieving roughly a 30% efficiency boost when refactoring a mini‑program to Taro. They also discussed model selection, embedding optimization, and the future where AI‑augmented developers outpace pure AI.

Baidi AI‑Native User Product Development

Fàn Zhōngkǎi compared past and present product‑user dynamics, defining AI‑Native as designing every feature with AI in mind while keeping AI invisible to end users. He linked AI‑Native concepts to Maslow’s hierarchy, covering content creation, quality assessment, distribution, and presentation. Practical advice for aspiring AI‑Native engineers includes model selection, context utilization, and prompt engineering.

JD’s Edge‑Intelligence Framework Exploration

JD showcased its edge‑intelligence system deployed in multiple JD App scenarios such as recommendation, search re‑ranking, and checkout risk control. The architecture spans cloud (model training and serving), edge (model management and distribution), and device (SDK‑based inference). Challenges discussed include computational performance, flexibility, stability, and security.

JD’s Cloud‑AI Collaborative Service Capability

Focusing on on‑device model computation, JD presented two business cases: search‑ranking re‑ordering and millisecond‑level risk control. The talk delved into model debugging, feature calculation, and inference pipelines, highlighting significant business gains from edge‑cloud collaboration.

JD’s “Tongtian Tower” Visual Page‑Building Platform

JD introduced the Tongtian Tower platform that enables visual construction of activity and channel pages across H5, native, and PC endpoints from a single codebase. The platform balances flexible customization with rapid page assembly, demonstrated through interactive mini‑games and lottery page examples, and outlines future directions such as unified front‑end design and AI‑driven generation.

Conclusion and Outlook

After three years, the JD & Baidu Frontend Salon aims to continue delivering high‑quality content, fostering industry exchange, accelerating technical innovation, and establishing standards that drive the frontend ecosystem forward.