The launch of ChatGPT in 2023 is regarded as a pivotal milestone in the history of human technological development. It swept the globe in just a few months at an unprecedented speed, driving the rapid advancement of technologies including AI algorithms, large models, and big data.

As a natural language processing (NLP) large model (referred to as "large language model") with outstanding capabilities in processing large-scale datasets, it reached new heights following ChatGPT’s emergence. The market witnessed a boom in diverse products and applications; by integrating with industries across the board, it explored viable commercialization models. Among these fields, open-source intelligence (OSINT) has undergone significant transformation and progress due to the introduction of large language models, forming a commercial closed loop through integration with artificial intelligence. A pioneer in exploring this industry model is Lu Siyuan—a doctoral student from the School of Electronic Science and Engineering at Nanjing University and founder of Nanjing Fengyu Intelligent Information Technology Co., Ltd. (hereinafter referred to as "Fengyu Intelligence").

Fengyu Intelligence is a company focused on open-source intelligence analysis using advanced artificial intelligence technology. Founded in September 2023, it has already secured two rounds of investment from Nanjing Innovation Investment Group and Miracle Plus.

"Since my junior year, I have been working with Professor Wang Zhongfeng—an international top expert in chip design and artificial intelligence—on research into next-generation AI technologies, particularly deep learning accelerator design. After being admitted to graduate school, I officially joined the Integrated Circuit and Intelligent Systems (ICAIS) Laboratory led by Professor Wang, focusing on speech recognition and natural language processing (NLP) technology. I was fortunate to be among the earliest researchers in China to focus on large model technology," Lu Siyuan expressed excitement and passion for the industry when introducing his academic journey. When Google groundbreakingly proposed the Transformer architecture—a neural network structure based on self-attention mechanisms—in 2017, Lu Siyuan realized through discussions with teachers and classmates in the laboratory that this architecture had transformative potential in the field of deep learning. Compared with traditional deep neural networks, Transformers are more efficient and concise; when integrated with industries such as healthcare, finance, and intelligence, they can further drive the development of related technologies and significantly improve information processing and decision-making efficiency.

As he continued to explore and research the technology, Lu Siyuan began to "take on the challenge" of developing hardware architectures that efficiently support Transformer model computing. He continuously balanced requirements in computing efficiency, memory access, energy efficiency ratio, and flexibility through R&D to adapt to the evolving trends of AI models and technologies. During his doctoral studies, he published his first paper on Transformer hardware architecture, which won the Best Paper Award at the 2020 IEEE International Symposium on System-on-Chip (SOCC). This achievement also made him one of the earliest researchers in the world to design a Transformer hardware architecture.

In November 2022, when ChatGPT—powered by GPT-3.5—emerged and gained widespread attention for its realistic natural language interaction and multi-scenario content generation capabilities, Lu Siyuan also grew more optimistic about the prospects of combining large models with commercial applications. He admitted, "I have always had the dream and idea of starting a business, and I hope to seize the opportunity of the rapid development of large model technology to turn my research into commercial value." With the support of Professor Wang Zhongfeng, he formed a team with Dr. Shi Yubo, who specializes in deep learning data compression. From three dimensions—system, algorithm, and hardware—the team addressed the challenges of deploying and implementing the latest deep learning NLP algorithms, aiming to independently design the RiSE architecture and NPU deployment toolchain while identifying competitive advantages and commercialization opportunities during the R&D process.

In March 2023, with the support of his supervisor Professor Wang Zhongfeng and the International Institute of Innovation and Entrepreneurship at Nanjing University, Fengyu Intelligence was officially established in the Nanjing University Student Entrepreneurship Park.

"The character ‘Feng’ (wind) symbolizes flow and rapid dissemination, while ‘Yu’ (language) is associated with speech and communication. Naming the company ‘Fengyu Intelligence’ reflects our hope that the company’s products can transmit and process information as efficiently and naturally as the wind. Therefore, we focused our efforts on R&D, but soon realized that startups inevitably face practical challenges related to funding, resources, and talent," Lu Siyuan noted. Although he is a serial entrepreneur, taking on the responsibility of leading the company’s development as a founder forced him to think more deeply and thoroughly. With the in-depth implementation of the "Ningju Plan" (a Nanjing initiative to attract and retain talents), more young entrepreneurs have chosen to stay and develop their businesses in Nanjing. To gain more support and attention, Lu Siyuan decided to lead his team in participating in the 2023 "Win in Nanjing" Youth College Student Innovation and Entrepreneurship Competition—hosted by the Nanjing Municipal Bureau of Human Resources and Social Security and supported by Nanjing Innovation Investment Group. After fierce competition, Lu Siyuan’s "Fengyu Intelligence" stood out among over 3,500 entrepreneurial projects, won unanimous recognition from the judges, and claimed the first prize in the competition. This achievement earned the company special fund support from the Angel Fund of Nanjing Innovation Investment Group and financial rewards. "Nanjing University provided us with significant support in terms of entrepreneurial venues and entrepreneurship education. Meanwhile, the angel investment and financial subsidies from the Venture Capital Group offered financial assistance, and their professional post-investment incubation services provided us with more professional guidance in understanding the relationship between products and the market, making our entrepreneurial journey more stable and reassuring," Lu said. In the same year, he submitted an application for an entrepreneurial project focusing on large language model applications to Miracle Plus—a well-known AI angel investment institution. The project gained the favor of Dr. Lu Qi, the founder of Miracle Plus, and was selected as one of the 60 projects in the 2023 Spring Entrepreneurship Camp, marking the first step in its market-oriented financing journey.

In April 2023, BesResearch released the 2023-2029 China Open-Source Intelligence Industry Market Development Trend Research Report. Covering industry overview, upstream and downstream ecosystems, leading enterprises, and core competitive products, the report highlights the significant role of open-source intelligence in supporting government and enterprise decision-making, strategy formulation, trend analysis, and risk prediction.

Lu Siyuan and his team 敏锐 ly captured the development prospects of the open-source intelligence industry, as well as the demand for efficient and accurate data acquisition, extraction, interaction, and rapid summarization of large volumes of text in the process of commercial intelligence detection, collection, and analysis. "Currently, the open-source intelligence industry continues to grow at a CAGR of 6.23%, and its market size is expected to reach 39.589 billion yuan by 2028. For the implementation of AI applications, this is a blue ocean market with enormous development potential and a high degree of alignment with the advantages of large language models. Therefore, we have made bold attempts in product R&D and market expansion," Lu explained.

Through in-depth analysis and strategic planning, Fengyu Intelligence focuses on the open-source intelligence industry. It leverages its expertise in high-performance computing for Transformer neural networks and deep research in deep learning data compression, fully utilizing the advantages of AI large models in text processing. By combining its strengths in large-scale pre-trained model compression algorithms and high-performance computing solutions, the company has refined its product capabilities and finally developed a "High-Performance Think Tank Agent" for vertical fields. This product integrates multiple advantages, including fast reasoning speed, high answer accuracy, rapid response, strong throughput, large external database capacity, interactive data governance, and private deployment capabilities. In the era of large models, the cost of acquiring and reusing training data has decreased, while efficiency has significantly improved.

"Compared with traditional products, the Think Tank Agent offers higher performance, faster response times, and better confidentiality. It features a RAG algorithm for multi-modal data (videos, images, text) and has no restrictions on the number of uploaded files. It can quickly analyze the content of uploaded files, enabling unified governance of various types of data and personalized information push. Only with such prominent advantages can we ensure a good user experience," Lu Siyuan introduced. "However, large model technology is costly, and deploying and implementing the latest deep learning NLP algorithms is particularly challenging. Our accumulated key technologies in NLP algorithms, hardware, cloud deployment, and embedded deployment give our products distinct advantages in both software and hardware, thereby significantly reducing costs." The high-performance Think Tank Agent is well-suited for application in the open-source intelligence field. By integrating artificial intelligence with open-source intelligence, Fengyu Intelligence has explored a new path for the implementation of large model products, driving the field toward an era of intelligence and automation.

Currently, the Fengyu Intelligence team has signed project cooperation agreements with 5 clients and is in discussions with over a dozen more, including leading government think tanks in Jiangsu, Bozhi Security, and Nanjing Cloud Data Technology. Among these collaborations, the development of a large model knowledge base for mental health consultations at a hospital is progressing effectively. By enhancing the hospital’s data analysis capabilities, the solution enables intelligent medical record generation, supports doctors in diagnosis and treatment, and improves both the speed and accuracy of consultations.

"On the entrepreneurial journey, the team is the helm; only through concerted efforts can we go far," Lu Siyuan stated. As the initiator, he formed a team with Dr. Shi Yubo (specializing in deep learning data compression) and invited academic and industry experts—including his supervisor Professor Wang Zhongfeng and Dean Pei Lei from the School of Information Management at Nanjing University—to serve as chief scientists and technical consultants. Backed by the ICAIS Laboratory at Nanjing University and its strong R&D background, the team collaborates closely with clear division of labor, drawing on large model technology to rapidly develop new products.

"Next, we hope to provide customers with integrated software-hardware solutions tailored to their practical application needs, achieving hardware-software integration for our products," Lu said when discussing future plans. He added that the company will continue to research and iterate on high-performance hardware equipment to provide critical guarantees for the security of Fengyu Intelligence’s products, ensuring that users’ internal data is not leaked when the product connects to the network to access external information. At the same time, the integration of software and hardware will strengthen the company’s product barriers and competitiveness.

Today, Fengyu Intelligence has developed hardware-software integrated products tailored to the specific needs of the intelligence industry, filed multiple intellectual property applications, secured project collaborations with several companies, and won numerous awards for outstanding projects. Behind these impressive achievements lies the team’s dedication and passion for entrepreneurship, as well as their keen insight into the commercial market and strong expansion capabilities.

The Mass Entrepreneurship Investment Team of Nanjing Innovation Investment Group stated, "We noticed Fengyu Intelligence through the ‘Win in Nanjing’ Youth College Student Innovation and Entrepreneurship Competition. As an innovative company in the open-source intelligence and artificial intelligence fields, this young entrepreneurial team from Nanjing University has a solid technical background, which supports the implementation of its products in the open-source intelligence sector. As a state-owned venture capital institution, we will continue to focus on the development of early-stage technology enterprises and provide the team with more post-investment support in terms of capital, management, resources, and talent. In the future, we will further strengthen cooperation with universities and research institutions in Nanjing, effectively implement the requirements for university-local integration development in Nanjing, support more young entrepreneurial enterprises like Fengyu Intelligence, and vigorously promote the transformation and implementation of more scientific research achievements by young university researchers."

"Through exploration, we have identified our target track and gradually realized a commercial closed loop from product R&D to market implementation," Lu Siyuan introduced. "Choosing to explore the application of large language models in the open-source intelligence industry is the result of careful consideration. In the context of great power competition, intelligence is the foundation of important decision-making, and technological strength is the most direct form of competitiveness. The lack of high-quality, timely, and accurate intelligence may put enterprises and even the country in an extremely passive position. Therefore, we also want to contribute to the great rejuvenation of the Chinese nation in the field we are good at—at least in the open-source intelligence track, helping the country build its core competitiveness!" Lu Siyuan’s eyes shone with determination when talking about his vision for deepening engagement in the open-source intelligence industry.