By Deke Wang and David Chen, Stream Computing
In the wave of digital transformation, the intelligent upgrade of community services has become key to improving the quality of life for residents. Stream Computing, a Chinese RISC-V intelligent computing solution provider, has created an integrated RISC-V hardware and software ecosystem for LLM applications through its self-developed RISC-V chips and inference acceleration card STCP920. This ecosystem includes an intelligent computing cloud platform, industry LLMs, a data governance platform, and an intelligent agent development platform. In particular, Stream Computing has developed an intelligent community service assistant agent for a community in Guangzhou, China, opening a new chapter in smart community services based on RISC-V.
Technical Solutions for Large Model Applications in Vertical Fields
As a chip enterprise with a comprehensive layout from the instruction set to the application of LLMs, Stream Computing has trained industry LLMs that are more suitable for community services using a large amount of vertical field data to support the deployment of large model applications on RISC-V computing power. In addition, to effectively suppress large model hallucinations, Stream Computing has developed a multi-round self-reflective Graph RAG algorithm, combined with knowledge graphs, to increase the accuracy of community Q&A to over 95%, making the deployment of large models in community services possible.
Smart Community Service Assistant Based on RISC-V Computing Power
Based on RISC-V computing power, Stream Computing has customized the development of a convenient community service assistant for a community in Guangzhou. The assistant covers 37 major categories of community services with over 2,000 service items, integrating thousands of documents and guidelines to provide residents with comprehensive intelligent Q&A services for community service consultation and handling processes. Compared to traditional service assistants, this service assistant has increased the accuracy of community service consultation from around 30% to over 95%, making the handling experience smoother and more natural, reducing the need for manual customer service, and promoting the accumulation of community service knowledge. Compared to traditional software systems, the intelligent agent can better serve business applications, allowing customers to not only obtain an intelligent Q&A system but also introduce a new way of working.
Deke Wang, Vice President of R&D at Stream Computing, stated: “Building a full-stack solution is a long and complex process that requires us to start from the perspective of computing as a service, designing reasonable boundaries for each product and function. Stream Computing always focuses on commercial implementation, whether it’s hardware or software products, aiming to develop end-to-end full-stack solutions right from the design phase.
Calista Redmond, CEO of RISC-V International, stated “Stream Computing’s contributions to the RISC-V ecosystem are impressive. By integrating RISC-V architecture with large AI models, they have demonstrated real-world application for advanced RISC-V technology. This is a forward-looking achievement that demonstrates the broad application prospects of the RISC-V architecture on a global scale.”
Outlook: Computing as a Service
With the successful launch of the STCP920 accelerator card, Stream Computing will continue to deepen its research and development in the RISC-V and AI fields. The company plans to further expand the functionality of its large model inference software stack and collaborate with more industry partners to jointly promote the intelligent development of smart communities and other vertical sectors. Additionally, Stream Computing will continue to focus on emerging technologies and market demands, continuously optimizing its products and services to meet the growing demand for high-performance computing.
This breakthrough by Stream Computing not only sets a new benchmark for Chinese enterprises in the high-performance computing and artificial intelligence sectors but also injects new vitality into the global technology industry. Through continuous technological innovation and market expansion, Stream Computing is steadily advancing toward its goal of becoming a leading global provider of computing technology.