Founded by researchers from the University of Toronto Machine Learning group and the Vector Institute, with ongoing academic and industry collaborations across UofT, Vector Institute, UC Berkeley, and Xiaohongshu — building small-model reasoning, agentic systems, and Chinese-language NLP that ships into production.
In collaboration with UofT · Vector Institute · UC Berkeley · Xiaohongshu
Jointly optimising reasoning and self-refinement under one binary reward — no critic model, no process reward, no critique data. Adds +11.5 pp on AIME pass@4 (Qwen3-4B) over vanilla GRPO.
50 software-engineering tasks drawn from a real production iOS codebase, paired with 449 human-verified test cases — a hard look at whether coding agents are ready for industrial mobile development.
A revolutionary multi-agent collaboration platform that intelligently decomposes tasks through a master coordinator agent, dynamically delegates to specialized sub-agents for Routed-Processing. Integrated with million-character document processing, multimodal search, local database integration, and other powerful features. Supports streaming responses to let users see AI's thinking process in real-time.
Built on the latest 网景自研架构, this next-generation Chinese grammar correction system employs multi-stage reinforcement learning training methods, achieving breakthrough improvements in false positive rates, processing speed, and deployment costs. Supports multimodal correction capabilities including text, video subtitles, audio transcription, and face detection, specifically designed for government, media, and education scenarios.
With 25 years of professional language monitoring experience, employing dual-architecture deep learning and rule-based engines for multi-dimensional correction, industry customization, and on-premises deployment.
Phishing and spoofed site detection, malicious redirect identification. Through knowledge graph validation and intelligent classification, achieving high-precision risk identification and remediation suggestions.
Simulating adversarial thinking to perform security assessments on intelligent systems. Supports on-premises deployment to ensure data security and comprehensively detect model vulnerabilities.
Image to structured data conversion. Combining optical recognition with intelligent validation, handling complex merged cells and low-resolution images, supporting financial reports, contracts, invoices, and more.
Coolwei AI Lab is an innovative team focused on cutting-edge research and applications in artificial intelligence. Our mission is to advance AI technology and create value for enterprises and society.