English
Back
Open Account

[Hardcore Recognition] 'Zhitu' Passes Huawei Ascend Certification, Shengtong Has More Plans for 2026

Have you ever encountered this situation:
Large models sometimes output very fluently, but upon verification, it becomes clear: factual inaccuracies, reasoning leaps, and unclear justifications. More importantly, when applied in high-demand scenarios like government affairs, finance, and healthcare, these kinds of 'hallucinations' are no longer a matter of user experience, but directly impact the credibility of implementation.
How can we make large models both articulate and accurate?
$VOICECOMM (02495.HK)$To what extent should 'trustworthy intelligence' be elevated?
In January 2026, Hubei Shengtong Intelligent Technology Co., Ltd.'s self-developed 'Zhitu' system officially passed the technical certification of Huawei Ascend AI's foundational software and hardware platform, receiving the 'Ascend Native Certification Certificate' and the right to use the official certification logo. The certification was jointly tested by Huawei Ascend Ecosystem Technical Certification Center and Hangzhou Intelligent Computing OpenLab, and validated based on the Ascend Atlas training server—indicating that Shengtong's 'Zhitu' has obtained a highly valuable 'hardcore pass' in terms of platform compatibility, core capabilities, and engineering implementation.
[Hardcore Recognition] 'Zhitu' Passes Huawei Ascend Certification, Shengtong Has More Plans for 2026
Localization Adaptation:This certificate also sends a more 'grounded' signal: Shengtong's 'Zhitu' platform can fully adapt to mainstream domestic AI computing power infrastructure, providing enterprise-level customers with a solid localized foundation for their applications.
Why was 'Zhitu' able to secure this authoritative endorsement?
Because Shengtong focuses not on 'making the model better at fabrication,' but on the more difficult and crucial task:Make the model more trustworthy
The 'ZhiTu' innovatively constructs a dual-engine architecture of 'large model + knowledge graph':
Knowledge graph 'anchors facts'
Industry knowledge, business rules, and historical data are structured and precipitated into a dynamic knowledge network, providing the large model with a traceable and constrained 'trustworthy coordinate system' to reduce deviations from the source.
Large model 'flexible generation'
With the computing power and stability support of the Ascend platform, the large model completes reasoning and generation under the guidance of the knowledge graph, maintaining smooth natural language interaction while being able to call structured knowledge more accurately, significantly reducing the probability of 'talking nonsense with a straight face'.
Integration: both smart and reliable
Through the deep integration of the two, the system response speed is significantly improved, and the efficiency of knowledge updating and iteration is greatly enhanced, truly achieving 'both smart and reliable'.
Behind this certification lies the recognition ofreliable operation and sustainable implementationwith dual recognition.
03 'SmartGraph' has already proven its 'hard power' in these scenarios
Currently, SoundConnect's 'SmartGraph' has been implemented in high-value scenarios such as government affairs, finance, healthcare, and industry:
Government affairs:Government Affairs: Policy interpretation and service guidance intelligent assistants rely on a government knowledge graph to provide responses that better align with regulations, reducing misleading information.
Finance:Credit risk control and compliance assistance, integrating legal regulations and credit reporting knowledge structures to help identify risk points and reduce misjudgments.
Healthcare:Medical record quality control and diagnostic support utilize a medical knowledge graph to constrain outputs, minimize omissions, and improve doctors’ efficiency.
Industry:Industry: Equipment fault diagnosis and quality inspection by linking operational data with maintenance knowledge systems to improve diagnostic accuracy and lower operation and maintenance costs.
These scenarios share only one common standard:Zero tolerance for errors, reliability is a must
2026: This recognition is just the beginning, more plans are on the way!
Obtaining the Ascend Native certification marks a key validation of Shengtong's technical roadmap. The company is expected to launch a brand-new intelligent agent platform in 2026—evolving from 'being able to answer' to 'being able to complete tasks and execute closed-loop operations,' advancing 'trusted intelligence' into a more usable, controllable, and scalable phase.
Stay tuned for more updates $VOICECOMM (02495.HK)$ . Next time, we will bring not only 'hardcore certifications' but also hope to introduce 'new products, new growth'”。
Risk Disclaimer: The above content only represents the author's view. It does not represent any position or investment advice of Futu. Futu makes no representation or warranty. Read more
1
5
+0
See Original
Report
56K Views
Comment
Sign in to view/post comments
avatar
Voicecomm
The official account of Voicecomm
7833
Followers
710
Visitors
Follow
Market Insights
HK Tech and Internet Stocks
View More
Nancy Pelosi Portfolio
Tariff game between the US and Europe shakes the market! Will TACO happen again?
Amidst the global market turbulence triggered by the US-Europe dispute over Greenland, a single post by Trump instantly reversed market tren Show More