Customers

Implementation playbooks for high-compliance scenarios

Government intranet meetings

No data leaves the site · MLPS Level 3

Scenario constraints

Meeting content covers internal deliberations; audio and text are explicitly barred from leaving the site
Procurement must satisfy MLPS Level 3 and domestic-technology catalogue requirements
Some sites are air-gapped, so models and dependencies must be delivered offline in one pass
Minutes must be retained long-term with search and audit support

Delivery model

VV10 deployed fully on-premise inside the government intranet; audio and text never leave the domain
Structured minutes + full-text search + compliant archiving to satisfy audit requirements
Minutes sync automatically to government OA platforms (Landray / Seeyon and others), with action items extracted and routed
For air-gapped sites, an offline bundle including CANN dependencies, model weights and Chinese fonts

Capability support

Serves nearly 100 government bodies including procuratorates, courts, police and cyberspace authorities
Meets MLPS Level 3 and classified-environment requirements
Headquartered in Suzhou with offices across East, Central, North, Northeast, Southwest, Northwest and South China, providing 24×7 service

View the matching industry solution →

Defence and research institutes

Air-gapped · domestic accelerators

Scenario constraints

The environment is fully air-gapped, so dependencies and licences cannot be pulled online
Procurement is restricted to domestic accelerators (Ascend / Cambricon / Hygon), requiring operator adaptation
Terminology is dense and general-purpose models underperform
Upgrades and maintenance must be completed offline

Delivery model

The speech engine natively supports Ascend 310P / 910B, Cambricon MLU370 / MLU590 and Hygon DCU, and also runs on CPU only
ONNX-to-OM model conversion with operator adaptation, plus an equivalence check between CPU and NPU outputs
Domain hotword injection and model fine-tuning to match specialist vocabulary
Offline upgrade bundles, with model conversion folded into CI so it tracks CANN version upgrades

Capability support

Undertakes national key projects including "Science and Technology Winter Olympics" and "Next-Generation AI 2030"
110+ invention patents (54 granted) and 200+ published papers (20+ CCF-A)
Won the 2016 Qian Weichang Award (First Prize) for Chinese Information Processing — the highest NLP award in China

View the matching industry solution →

Financial-sector meetings

Regulatory traceability · research terminology

Scenario constraints

Research meetings contain non-public information and cannot be processed on third-party clouds
Regulation requires meeting records to be traceable and auditable
Financial terminology and product names are dense, so generic recognition is error-prone
Must integrate with existing research, risk-control and OA systems

Delivery model

On-premise deployment with zero external connectivity, meeting regulatory data-residency requirements
Structured minutes with retained decisions and action items form a complete auditable trail
Financial-sector hotword dictionaries and domain fine-tuning to match business terminology
Standard APIs integrate with research, risk-control and OA systems so minutes land in the existing workflow

Capability support

Partnerships with 200+ industry leaders including iFlytek, Tencent, Huawei and Kingsoft
Meets MLPS Level 3 and classified-environment requirements with zero data egress
Deep integration with WeCom, DingTalk, Feishu, Landray and Seeyon

View the matching industry solution →

Clinical consultation and academic meetings

Terminology precision · patient data stays in-hospital

Scenario constraints

Patient-related information is protected by privacy rules and cannot be sent to the cloud
Drug names, procedures and specialty terms are highly technical, so generic models produce many errors
Consultations involve multiple specialists speaking in turn, requiring speaker separation
Must integrate with in-hospital HIS / OA systems

Delivery model

In-hospital on-premise deployment so patient data never leaves the campus
Medical hotword dictionaries plus ASR × OCR fusion: on-screen case and imaging text reinforces recognition
Speaker diarization automatically separates participants and produces per-speaker consultation minutes
Integration with in-hospital HIS and OA so minutes enter the existing workflow

Capability support

ASR × OCR fusion lifts terminology recognition from about 72% to 94%+ (measured by the NEU NLP Lab)
Recognition accuracy is 98% or higher for standard Mandarin and 96% or higher for dialects, across 22 dialects
52 years of NLP heritage since 1973, from the NLP Lab, Department of AI, Northeastern University

View the matching industry solution →

Multinational enterprise and education

30 languages + 22 dialects · real-time

Scenario constraints

Multinational meetings mix languages, and Chinese-English code-switching must work without manual selection
Accents vary widely (Southeast Asia, South Asia, the Middle East, Latin America and more)
Teaching settings are dense with specialist terminology and rely on the projected screen
Meeting and course content involves trade secrets or unpublished research

Delivery model

Recognition across 30 languages with LLM translation and summarisation for 100 languages, and automatic detection of 22 dialects
Targeted retraining for regional accents; accented Mandarin and accented English both measure above 90%
ASR × OCR fusion: BJ66 + WCB06 dual-channel screen capture with on-device large-model parsing and hotword injection
The pipeline runs locally; audio and text never traverse the public internet

Capability support

Recognition covers 30 languages plus 22 dialects, with LLM translation and summarisation for 100 languages
BJ66 with WCB06 captures the projected screen over wired HDMI and wireless dual channels
200+ discipline knowledge bases hot-swappable, with hotword injection latency under 500 ms

View the matching industry solution →

This page describes solution models and delivery capabilities, not customer names. All capability figures come from publicly available information on this website.

Book a Personalized Demo

Tell us your meeting scenario and compliance needs — get a tailored plan.

Book now