Before "large model" (大模型) became a household phrase in China, one engineer had already spent fifteen years trying to make machines understand language. His name rarely appears in consumer ads, yet his work sits underneath Baidu's search box, its translator, and its ERNIE (文心) models.
Wang Haifeng (王海峰) is Baidu's executive vice president and chief technology officer — and, quietly, one of the most influential architects of China's domestic AI software stack.
Key takeaways
- Wang Haifeng (王海峰) is Baidu's executive vice president and CTO, and director of the National Engineering Laboratory for Deep Learning Technology and Applications (深度学习技术及应用国家工程研究中心).
- In 2013 he became the first Chinese president of the Association for Computational Linguistics (ACL) in its 50-plus-year history, and was elected an ACL Fellow in 2016 — the first mainland China scientist to earn that honor.
- After joining Baidu in 2010, he built the company's technology directions in NLP, knowledge graphs, speech, vision, machine learning and deep learning.
- He led PaddlePaddle (飞桨), China's first open-source industrial-grade deep-learning platform, and the ERNIE (文心) knowledge-enhanced large model (大模型).
The translator who became an architect
Wang's route to the top of a technology company ran through universities and labs rather than the business side. He earned his bachelor's, master's and doctoral degrees in computer science from Harbin Institute of Technology (哈尔滨工业大学) between the early 1990s and 1999, with a research focus on machine translation and language technology.
Before Baidu he worked at Microsoft Research Asia and then spent roughly eight years as chief scientist at Toshiba's China R&D Center, leading work on natural-language processing, machine translation and speech. That stint taught him the gap between a published algorithm and a product millions of people actually use — a lesson he carried into Baidu when he joined in 2010.
Building the stack, layer by layer
Inside Baidu, Wang's signature move was to turn language technology from a scattered set of features into a platform. He established a dedicated NLP department and, over the following years, the knowledge-graph, speech, vision, machine-learning and deep-learning directions that now underpin the company's products.
Two results stand out. The first is Baidu Translate, which his teams built into a service supporting 203 languages — a concrete, measurable footprint that reflects two decades of steady translation research rather than a sudden breakthrough. The second is the knowledge graph his groups assembled, among the largest of its kind, used across search and recommendation.
PaddlePaddle and ERNIE
Wang's most cited legacy is PaddlePaddle (飞桨), which Baidu describes as China's first open-source, fully featured industrial-grade deep-learning platform. Positioned as the "operating system of the intelligent era," it sits beneath models, frameworks and applications, giving Chinese developers a domestically sourced alternative to foreign toolchains.
On top of that foundation sits ERNIE (文心), the knowledge-enhanced large model (大模型) family that became Baidu's answer to the global foundation-model wave. Wang's role was less to author a single paper than to keep the layers — data, framework, model, application — advancing together so each could reuse the last.
Recognition, and the limits of it
Wang has collected the kind of honors that mark a senior figure in Chinese engineering: a National Science and Technology Progress Award (Second Prize), the Guanghua Engineering Science and Technology Award, the first Wu Wenjun AI Outstanding Contribution Award (吴文俊人工智能杰出贡献奖), and the "National Outstanding Engineer" (国家卓越工程师) title awarded in 2024.
It is worth stating plainly what such a title does and does not mean. It recognizes sustained technical and industrial contribution; it is not an independent verdict on any single product's quality. His decades-spanning role across both academia and Baidu makes him an unusual figure who can speak the language of peer-reviewed research and product roadmaps alike. Readers should treat the honors as evidence of career impact, not as a scoreboard for model benchmarks.
What readers can do now
- Try the domestic stack directly. If you build with Chinese AI tools, PaddlePaddle (飞桨) and ERNIE (文心) are worth a hands-on test as an alternative to foreign frameworks.
- Judge model claims by independent benchmarks. Company comparisons of ERNIE to overseas models are largely Baidu's own statements; verify with third-party evaluations before relying on them.
- Study the career shape. Wang's path — deep academia plus long industrial tenure — shows natural-language processing as a long-horizon field where compounding experience matters more than a single headline result.
Honest limitations
This profile is based on Baidu's official biographies, state-media reports (China Economic Net, China Securities Journal) and Wang's own professional page. Some ecosystem numbers, such as developer or enterprise counts, are company-reported and not independently audited; we have therefore kept them qualitative or limited to well-sourced figures like the 203-language translator. Capability comparisons between ERNIE and foreign models are presented as Baidu's claims, not established fact. The focus is Wang's verified role and milestones; internal strategy and unpublished work are outside scope, and no product or security is endorsed.