The Genesis AI Innovation Center of Futong Technology partnered with Sapient Intelligence to upgrade its Voice of Customer (VoC) intent recognition capabilities based on the open-source HRM-Text-1B ultra-lightweight model. In validation across Futong Technology’s core automotive business scenarios, the upgraded model achieved over 90% overall accuracy in complex intent recognition tasks, demonstrating stronger stability and generalization capabilities in challenging scenarios such as multi-intent expressions, semantic ambiguity, and automotive-specific terminology recognition.
A More Automotive-Aware VoC Model for Complex Intent Recognition
Within the digital operations systems of the automotive industry, VoC platforms have become a critical bridge connecting customer experience insights with business decision-making. Futong Technology has accumulated more than eight years of expertise in automotive VoC solutions, enabling unified ingestion and governance of multi-modal unstructured data including voice, text, and images. On top of this foundation, the platform transforms raw customer feedback into structured semantic insights and business-oriented intelligence, empowering automotive enterprises across the full value chain, from product R&D and marketing to sales, after-sales service, and operations.
The Futong VoC platform integrates multiple industry AI models, including sentiment analysis models, opinion extraction models, tag classification models, and intent recognition models, enabling a comprehensive understanding of customer feedback.
Through this joint model fine-tuning collaboration with Sapient Intelligence, Futong Technology further strengthened the system’s ability to understand historically challenging scenarios, including intertwined multi-intent expressions, semantically ambiguous statements, and deeply implicit signals.
Across multiple rounds of validation in core business scenarios, the upgraded model achieved an overall accuracy rate exceeding 90% on complex intent recognition tasks, representing a significant improvement over previous-generation models under comparable conditions.
The upgraded model can now be integrated into Futong Technology’s VoC processing workflow, improving the efficiency and quality of transforming unstructured customer feedback into actionable business intelligence. The model outputs can be directly mapped into business processes and operational decision-making systems, supporting applications across marketing, sales, customer service, and product operations.
Lightweight Open-Source AI, Fine-Tuned for Automotive VoC Scenarios
This upgrade is based on Sapient Intelligence’s open-source HRM-Text-1B ultra-lightweight reasoning model, officially released as open source by Sapient Intelligence on May 19, 2026. This model incorporates three core innovations: task-oriented learning, latent-space reasoning, and a hierarchical recurrent architecture. Using only approximately 40 billion tokens - compared with the trillions of tokens typically required by conventional large models - and roughly $1,000 in pretraining costs, HRM-Text achieved reasoning performance on benchmarks such as MATH and DROP that surpassed traditional models 2–7 times larger in parameter scale.
The model contains only 1 billion parameters and requires just 0.6 GiB after int4 quantization, enabling fully localized deployment while ensuring sensitive customer data remains within private environments and meeting enterprise requirements for data residency and privacy compliance.
Futong Technology injected its proprietary automotive VoC annotated datasets and task frameworks into the model through post-training fine-tuning tailored to industry scenarios. This adaptation process enabled the model to evolve from a general-purpose reasoning capability into a domain-specific engine optimized for complex enterprise tasks, including multi-intent recognition, joint sentiment-and-intent analysis, and risk signal extraction.
Without relying on cloud-based frontier models or massive compute infrastructure, the upgraded solution is able to deliver practical, high-accuracy intent recognition capabilities directly within real-world business scenarios.
Continuing the Exploration of Lightweight, Efficient AI for Vertical Industries
As an important platform for AI innovation and engineering capability development within Futong Technology, the Genesis AI Innovation Center will continue driving the implementation and evolution of AI technologies across the company’s product portfolio, while strengthening the synergy between data intelligence, model capabilities, and industry applications.
Looking ahead, Futong Technology will continue promoting the deep integration of AI models, industry knowledge, and business scenarios, exploring more efficient, lightweight, and practically deployable AI pathways for vertical industries. The company aims to bring AI capabilities directly into operational workflows and business decision-making processes, continuously unlocking both data value and business value.
