Technology
Japan
SME/Startup
“Alibaba Cloud's technical support engineers provide accurate answers to even the most minor implementation queries. Even when I'm unsure about the scattered information available online, I feel at ease knowing that I can ask the developer directly. "
Gaku Kamimura
CEO/CTO, TIGEREYE, Inc.
About TIGEREYE
TIGEREYE, Inc. is developing a multimodal AI platform designed to address social challenges leveraging AI's "eyes" and "ears", adhering to its vision of "from recognition to communication". The company provides AI agents that can "understand and communicate" in real time by integrating recognition-based AI, such as facial recognition, skeleton estimation, emotion analysis, speech understanding, and natural language processing (NLP), with large language models (LLMs) and speech synthesis technology. ▶ Watch the full interview
Challenge
The biggest challenge was providing Japanese language support. Specifically, TIGEREYE's unique prompt rule design and LoRA fine-tuning were essential for accurately interpreting Japanese layout during optical character recognition (OCR) and managing the mix of honorific and colloquial language in Office Agent.
Why Alibaba Cloud
Alibaba Cloud was chosen for its processing power capability to run LLMs and a reliable support system. Close collaboration with Alibaba Cloud's technical support engineers enabled quick confirmation and responses to inquiries not addressed in manuals or online resources, such as specific sizing details.
Key Results
The following outcomes were attained by calibrating Qwen as a "local LLM" and leveraging high-performance processing capability for accelerated learning. ● TTS/Voice Cloning: Reduced learning time by 25%, significantly improving the naturalness of synthesized speech.
● Dual OCR: Improved form classification accuracy by 12%. This capability is currently being tested and implemented in local public service sectors and medical institutions.
● Office Agent LLM: Increased dialogue success rate to over 87% in internal business support scenarios, such as Slack integration and application assistance.
Looking Forward
TIGEREYE aims to continue enhancing Japanese language support across models, data, and evaluation. It strives to optimize the use of computing resources by advancing GPU schedulers and reservation control. Additionally, it will focus on optimizing Qwen's RAG model and expanding voice input/output features to enhance the product's completeness. Furthermore, the company aims to strengthen its collaboration with the marketing and Go-to-Market strategies to commercialize generative AI. TIGEREYE looks forward to deepening its partnership with Alibaba Cloud and working together to shape this emerging industry.
Featured Products
Alibaba Cloud provides Tongyi Qianwen (Qwen) models, a series of large language models (LLMs) and multimodal models (MLLMs), to the open-source community.
Other Related Stories
CTW
By leveraging Alibaba Cloud services, CTW's G123 platform now delivers high availability, low latency, and automated resource scheduling, significantly improving stability, load capacity, and cost efficiency.
BAEL
Alibaba Cloud helped BAEL reduce monthly infrastructure costs by 20%-30% while ensuring stable and efficient game operations through advanced cloud solutions tailored to handle surges in user traffic.
Mynet
Alibaba Cloud helped Mynet to reduce infrastructure costs, enhance performance with advanced technologies, and ensure world-class security for their gaming products, enabling the company to optimize operations and deliver a better gaming experience.
Snapshot
With Alibaba Cloud's comprehensive technical support, TIGEREYE is accelerating the development of conversational AI agents for the Japanese market.
Product/Solution Used
View More Solutions
Related Whitepaper
Alibaba Cloud Generative AI Solution Whitepaper
This whitepaper is a comprehensive guide on the generative AI (GenAI) products, technology, and capabilities of Alibaba Cloud.
Download
A Free Trial That Lets You Build Big!
Start building with 80+ products and up to 12 months usage for Elastic Compute Service
Get Started for Free Get Started for Free