DeepSeek disrupts industry landscape, Unity Ventures portfolio companies embrace new AI era | Unity Ventures portfolio

九合创投九合创投··0·4

Move fast and seize the initiative.

We are witnessing a "DeepSeek moment."

DeepSeek has rewritten the rules of the AI industry and is building an entirely new ecosystem.

On the new wave sparked by DeepSeek, we believe:

  • DeepSeek has significantly lowered the barrier to developing large model applications, especially for applications requiring strong reasoning capabilities, opening up a broader market. Technological progress has pushed the application layer to an inflection point in consumer awareness.

  • Unity Ventures continues to be bullish on AI applications that meet specific needs in concrete scenarios. We believe that improved inference efficiency and the marginal cost reduction brought by open-source models will further accelerate the adoption of generative AI.

  • An explosion of applications is imminent. Downstream from large models, entrepreneurial teams that focus on product refinement and possess strong product capabilities will enter a golden age of development.

Unity Ventures has long been deeply rooted in the AI field, with its portfolio of AI companies spanning both the technology layer and the application layer. From cloud computing providers to entertainment assistants, smart hardware, and intelligent customer service, Unity Ventures-backed companies are exploring the most commercially valuable coordinates in this transformation through real-world deployment.

HakkoAI Integrates Full-Power DeepSeek-R1, Upgrading Based on Player Needs

HakkoAI is an AI companion for gamers, providing persistent real-time voice chat, visual recognition-based proactive recommendations, and other services through a desktop pet companion experience. It has been widely acclaimed by the industry and players, having served 6 million players to date, and was honored as an Outstanding AI Product of the Year in 2024. Its parent company LynkSoul ranked fourth on The Information's list of most promising global startups.

HakkoAI was among the first to integrate the full-power DeepSeek-R1 after the holiday, leveraging its enhanced reasoning capabilities to better serve gamers across scenarios including finding games, playing games, and reviewing games. The platform is also expanding into additional scenarios that young people need, such as learning, video companionship, and personal assistance.

Meanwhile, the DeepSeek-Distill-Local 1.5B model has shown tremendous potential in helping HakkoAI with player intent recognition and ASR accuracy improvement during real-time in-game dialogue. This will be rolled out progressively to better serve gamers using large language models in real time during gameplay.

Yunxigu Integrates DeepSeek-R1, Ushering in a Leap for Smart Terminal Products

As an innovator in smart learning terminals, Yunxigu has actively embraced the DeepSeek technology wave. Its flagship products, including smart alarm clocks and smart dictionary pens, have undergone comprehensive algorithm upgrades, achieving breakthrough progress in core functions such as problem scanning and intelligent chat companionship, accelerating the innovation cycle of smart terminal products.

After integrating the DeepSeek-R1 large model, Yunxigu's smart alarm clock has seen a revolution in wake-up interaction. It can now understand more complex natural language commands, implement more personalized customized reminders, and provide more humanized wake-up suggestions based on user sleep data and schedules, making every morning smarter and more thoughtful.

The smart dictionary pen, empowered by DeepSeek-R1, has significantly improved scanning recognition speed and definition accuracy, with notably enhanced parsing capabilities for rare characters and complex sentence patterns, doubling learning efficiency. Building on DeepSeek-R1's powerful reasoning capabilities, Yunxigu has innovatively optimized its problem-scanning algorithm, enabling more precise identification of various question types and generating clearer, more accessible, multi-angle problem-solving approaches, truly achieving "one pen in hand, no difficult problem too tough."

Additionally, Yunxigu is exploring the application of DeepSeek-R1 in chat companionship functions for smart terminals, striving to create a more fluid, natural, and emotionally nuanced AI companion that provides users with warmer, more personalized interactive experiences.

Airdoc Wanyu Medical Large Model Upgraded

On February 7, Airdoc's self-developed Wanyu medical large model completed its upgrade and integrated the DeepSeek R1 model.

Airdoc Wanyu helps medical professionals quickly access medical information, clinical guidelines, and research findings to improve decision-making efficiency. It also provides patients with personalized health condition analysis and treatment recommendations. Previously, the Wanyu medical large model had already been commercially deployed in medical examination scenarios.

Airdoc's multimodal analysis system has achieved three advances. First, it can more intelligently and efficiently predict risks for over 50 diseases through a single fundus photo. Second, the system innovatively integrates evidence-based medical knowledge to generate structured reports covering lesion localization, concurrent risk warnings, and other dimensions, improving physician decision-making efficiency by 30%. The doctor-patient interaction module uses natural language generation technology to provide personalized explanations of treatment plans, increasing health guidance compliance by 65%.

In the AI myopia phototherapy field, the Wanyu medical large model has transformed myopia phototherapy devices from single-function treatment tools into highly autonomous "home vision care specialists." Through precise tracking of axial length changes, it generates structured visual health reports and provides data-based personalized intervention recommendations.

In AI visual training scenarios, the Wanyu medical large model further enhances the intelligence of AI eye-movement visual training products, significantly improving physicians' auxiliary capabilities during diagnosis and treatment processes.

Tungee Integrates DeepSeek Large Models to Drive Intelligent Enterprise Growth

As an AI service provider focused on sales, Tungee has integrated and deployed the DeepSeek series of large models.

Tungee's leading data intelligence technology platform for enterprise intelligent operations was among the first to access DeepSeek's AI application modules, building intelligent analysis, intelligent Q&A, and other intelligent modules to create an all-purpose AI assistant for enterprise intelligent business operations.

The platform's intelligent analysis module can deeply analyze core business data metrics and dynamically generate analysis reports automatically. Based on DeepSeek's MoE (Mixture of Experts) architecture and dynamic learning rate optimization technology, it can real-time parse fluctuations in core metrics such as enterprise performance, renewal rates, and conversion rates, predict trends based on market dynamics monitoring and historical data, and provide recommendations for production volume, inventory reserves, and logistics transportation, generating analysis reports on a regular basis.

Additionally, the platform features an intelligent Q&A module that leverages DeepSeek's deep reasoning and generation capabilities to achieve rapid understanding, deep organization, and precise retrieval of knowledge based on industry knowledge graphs, providing explainable decision recommendations through natural language rather than simple answers. Tungee's intelligent Q&A module supports multimodal inquiry methods including text, voice, images, and tables, with an accuracy rate exceeding 98%, serving as an enterprise-exclusive Q&A intelligent agent.

Tungee's other AI product applications, including Tungee SalesGPT and Tungee SalesGBN, have also fully integrated the DeepSeek series of large models. Tungee SalesGBN combines DeepSeek-R1's exceptional logical deduction and concurrent reasoning capabilities with a comprehensive knowledge graph of 180 million enterprises, using industry-specific professional knowledge (Know-How) as the core training source to inject expert knowledge into the large model. This enables it to not only understand general sales scenarios but also deeply adapt to vertical industry-specific needs, helping enterprises rapidly expand new customer bases and break through traditional customer acquisition cognitive barriers.

Xiaoduo Integrates DeepSeek-R1 Large Model, Combining Forces with Self-Developed XPT Model

As a leader in intelligent customer service, Xiaoduo Technology has fully embraced an AI strategy. Currently, Xiaoduo has integrated both DeepSeek-R1 and DeepSeek-V3 large models, committed to incorporating cutting-edge AI technology into products including Bisheng Robot, AI Training Ground, VOC, Aggregated Reception, and Keban, to provide intelligent services for customers.

The AI intelligent review response feature jointly developed by Xiaoduo Technology and Bazhuayu, after integrating DeepSeek, can more precisely identify buyer sentiment and review key points, achieving a qualitative leap in automatic review responses and providing merchants with more efficient, precise, and warmer review management solutions.

Meanwhile, Xiaoduo's AI intelligent review response function has further expanded its supported platform scope, covering more mainstream e-commerce platforms including Tmall, JD.com, and Douyin. Merchants no longer need to switch between different platforms; through this feature, they can achieve one-click batch automatic review responses, processing large volumes of reviews at once, significantly saving time and labor costs. Previously tedious review reply work has become effortless and efficient, allowing merchants to devote more energy to other important aspects of store operations.

TTC Upgrades Product AI Core, Enhancing Enterprise Recruitment Efficiency

TTC (True Talents for China), as a leading AI-driven recruitment platform in China, has consistently dedicated itself to improving enterprise recruitment efficiency through technological innovation.

As early as May 2024, TTC had already integrated DeepSeek V2 into its business workflow to perform tasks such as resume parsing, tagging, and person-job matching analysis, improving information processing accuracy while significantly reducing computing costs.

By the end of 2024, TTC upgraded the AI core of its AI product "Xiaomai" to DeepSeek V3, substantially enhancing its intelligence level and delivering outstanding performance in multi-scenario recruitment, earning widespread acclaim.

In January 2025, TTC further upgraded to DeepSeek R1, expanding into more internal collaboration and management application scenarios, further improving the intelligence and efficiency of the recruitment process, and providing more precise and efficient talent matching services for clients in industries including AI, Embodied Artificial Intelligence, robotics, new energy, high-end intelligent manufacturing, biomedicine, and new consumer goods.

Going forward, TTC will continue to leverage the most advanced AI large models, continuously optimizing the talent service platform where AI Agent "Xiaomai" collaborates with professional human consultants, constantly accumulating and refining professional service data, and using an AI + human model to provide smarter and more efficient recruitment services for clients and candidates.

Xingzhe AI Integrates DeepSeek into Product Ecosystem, Empowering Digital Entertainment and Multi-Domain Intelligent Transformation

Xingzhe AI is dedicated to exploring and mining application scenarios in artificial intelligence, with a focus on digital entertainment, digital cultural tourism, and digital education. As early as the end of 2024, Xingzhe AI began paying attention to DeepSeek, this domestic open-source large model, and rapidly initiated testing and model training.

Today, Xingzhe AI has completed model training based on open-source DeepSeek combined with its own data, and achieved deep integration across multiple core products with significant results.

This deep integration covers multiple core products under Xingzhe AI, including multimodal vertical large models for digital entertainment and smart education product series. Through comprehensively enhancing large model adaptation capabilities, the interactive experience of Xingzhe AI's smart education product series has been notably improved.

In terms of technology R&D, DeepSeek's code generation, debugging, and optimization functions have provided powerful support for Xingzhe AI's R&D team, significantly improving development efficiency. In terms of technology strategy, Xingzhe AI's "Multimodal Vertical Large Model for Digital Entertainment Industry" is built upon accumulated hundreds of billions of data points, self-developed art vertical models, music vertical models, and computing infrastructure. By adopting the same distillation logic as DeepSeek-R1 to further optimize structural models, it improves model operational efficiency, reasoning capabilities, and generalization capabilities, providing more comprehensive content creation support for the entertainment industry.

At the product level, Xingzhe AI's smart education product series has been upgraded through DeepSeek's natural language processing and reinforcement learning technologies, achieving more precise personalized learning, cross-scenario and cross-disciplinary linked teaching, intelligent teaching assistance, and more natural interactive teaching, bringing users a brand-new digital teaching and learning experience.

Jingying Technology Integrates DeepSeek Large Model, Giving AI Short Drama Creation Wings

CreativeFitting (Jingying Technology) aims to become the world's largest AIGC content platform and is currently a pioneer in the AI short drama industry. Its AI short drama creation platform has reached industry-leading standards in narrative video generation capabilities. The company also operates the world's first AI short drama app "Reel.AI," completing the closed loop of AI short drama commercial distribution and bringing users an immersive AI-native entertainment experience.

CreativeFitting's AI short drama creation platform has integrated DeepSeek, allowing directors to directly use natural language to describe creative intentions and make video adjustments. The large model transforms these into nuanced, professional shot descriptions, and through intelligent tools automatically completes tedious details such as shot scheduling, character setting, and scene construction. The creation process becomes more flexible and fluid, with creative potential further unleashed.

Currently, numerous directors have produced and distributed multiple high-quality short dramas through CreativeFitting's AI short drama creation platform, landing on overseas trending charts. CreativeFitting continues to push the boundaries of AI content and experience, consistently leading transformation and innovation in the AI + entertainment space.

Image source: CreativeFitting-distributed AI short drama for overseas markets

QingCloud Launches Full DeepSeek Large Model Series

QingCloud's AI computing cloud service — CoresHub — has deployed and launched the full DeepSeek large model series for users to choose from, including DeepSeek R1/V3's 671B full-power version, quantized versions, and the full-size DeepSeek R1 Distill distilled model series. It supports multiple model delivery methods to meet different users' personalized needs regarding data security and model usage.

Novice users can directly ask questions and converse by calling the CoresHub DeepSeek-R1 API through various AI applications or chat clients. Intelligent knowledge base RAG and AI agent developers can call the CoresHub DeepSeek-R1 API for development through RAG/AI Agent platforms. Intelligent application developers can directly call the CoresHub DeepSeek-R1 API for application development. Researchers can use the CoresHub platform and AI lab tools for model distillation and fine-tuning.

Additionally, CoresHub has launched enterprise knowledge base private deployment services based on DeepSeek, which can adapt different DeepSeek versions and knowledge base components according to enterprises' personalized needs, achieving out-of-the-box, rapid delivery without excessive professional operations, providing enterprises with one-stop knowledge base construction services.