Building AI‑Native Applications with Baidu Cloud AppBuilder

This article summarizes Sun Ke’s keynote “Building Applications in the AI‑Native Era” delivered at the 2023 Baidu Cloud Intelligence Conference. The speech outlines how the AI landscape has shifted from model selection and evaluation to the practical challenge of building AI‑native applications.

Key challenges highlighted include rapid model selection, prompt engineering for complex tasks, controlling hallucinations, and efficiently locating suitable application prototypes. The speaker emphasizes the need for engineering solutions that bridge large‑model capabilities with real‑world application requirements.

The Baidu Cloud AppBuilder platform is introduced as a comprehensive solution. Its architecture consists of three layers:

Multimodal capability components (e.g., ASR, TTS, text‑to‑image) that, while not always LLM‑driven, are essential for AI‑native app development.

LLM‑based capability components such as NL2SQL, which encapsulate prompt engineering into reusable services.

Basic infrastructure components (cloud storage, vector databases, hardware‑software integrated multimodal modules) that provide a one‑stop development environment.

Developers can create applications using two modes:

Low‑code mode: Point‑and‑click configuration to quickly generate an AI‑native app, then debug and integrate it into existing systems.

Code mode: Full SDK, example code, and debugging tools are provided. All example code is open‑source, allowing developers to customize or replace components as needed.

Two concrete examples illustrate the platform:

HR resume screening: An LLM parses job descriptions, extracts keywords, and uses a RAG framework to match resumes automatically.

Interview preparation: An LLM summarizes resumes, highlights key points, and suggests interview questions, demonstrated in both low‑code and code modes.

The talk also showcases additional capabilities such as the Baidu ElasticSearch (BES) vector search engine, intelligent data analysis (upload‑Excel → instant query), and various generative tools (style‑marketing copy, Q&A pair generation, text tagging). These services are offered via APIs/SDKs and can be integrated into custom workflows.

Overall, the presentation positions AppBuilder as an AI‑native development accelerator, enabling developers to rapidly prototype, customize, and deploy AI‑driven solutions across industries.