Unlocking Agile AI Development: Insights from Huawei’s AI Native Apps Summit

On June 15, the HCDG City Tour Beijing stop – “Exploring and Innovating AI‑Native Applications” – was successfully held, bringing together experts from Huawei Cloud and the Huawei Cloud HCDE community to empower developers and enterprises pursuing AI innovation.

Agile AI Workflow

The event opened with a keynote titled “Agile Development of AI Workflows” by Gao Yuan, co‑founder & CTO of Sentence Interactive and Huawei Cloud HCDE. He emphasized that agile development relies on rapid iteration and continuous delivery, requiring constant feedback collection and swift strategy adjustments to meet evolving requirements and technological contexts. He identified functional challenges such as precise memory control, blended proactive‑reactive communication scenarios, and integration of proprietary system logic, as well as iteration challenges like analyzing unexpected responses and ensuring new changes do not disrupt existing logic.

Gao summarized three agile strategies: (1) fast iteration needs fast feedback, achieved via a tuning center and one‑click test import; (2) node verification and post‑execution validation ensure clear, visible results; (3) version control isolates testing and production environments, with cross‑version test cases for effect comparison. Live demonstrations deepened developers’ understanding.

Huawei Cloud’s Ascend Computing and the Pangu Model

Solution expert Zhang Ruifeng introduced Huawei Cloud’s Ascend AI chips and the Pangu large model, highlighting the massive potential of the large‑model industry as a digitization of human knowledge. He noted that building foundational models requires sustained investment from enterprises and talent. Leveraging Huawei’s end‑to‑end AI solution stack, developers can fully exploit AI compute power through the ModelArts one‑stop development platform, enabling rapid training, inference, and deployment.

Developers can also deploy open‑source models on AI Gallery with one‑click deployment for hundreds of model variations. Zhang explained that large‑model development is a complex systems engineering effort, demanding expertise in large‑scale data processing and distributed parallel training, covering cluster construction, data preparation, model training, inference compression, and application development. Huawei Cloud provides a complete toolkit to address these challenges, boosting industry‑wide model development efficiency and creating value for users and partners. He also showcased Pangu model applications across government, internet, film, and gaming sectors, illustrating AI’s transition from niche exploration to pervasive industry adoption.

Human‑Like AI Companions

Finally, Zheng Shuliang, CEO of Lingxin Intelligence and Huawei Cloud HCDE, presented “Super‑Human‑like Models and Psychological Support & Companionship.” He described super‑human‑like models as extensions of efficiency and creative value that also deliver emotional value, serving personalized needs in emotional companionship, mental health, and education. He analyzed models such as Emohaa, CharacterGLM, and OPD‑2 from perspectives of high security, strong controllability, and anthropomorphism, and discussed their deployment in education, entertainment, and marketing to provide AI‑native social and companionship experiences with immersive multimodal interaction.

The event also featured Huawei Cloud DTSE experts (developer technical support engineers) who engaged with developers to solve technical problems and provide cloud development guidance. Huawei Cloud reaffirmed its commitment to collaborate with developers to advance AI large‑model innovation and build a thriving AI ecosystem.