TranscEngram (Yisheng Technology) - 忆生科技

TranscEngram, operating publicly under the brand name Engram and also referred to as Yisheng Technology, is a China-based startup developing embodied AI systems intended to give robots memory-based, brain-like control architectures rather than purely reactive control loops. The company describes its mission as moving from what it calls 'AI 1.0' (task-specific artificial intelligence) toward 'AI 2.0' (autonomous intelligence that learns continuously from open, real-world physical environments), and its work is closely tied to research groups at the University of Hong Kong, including HKU's School of Computing and Data Science, the HKU Shanghai Intelligent Computing Research Institute, and the HKU Autonomous Intelligence Lab. Its named technical projects include CRATE, a white-box, mathematically interpretable deep neural network architecture that the company states was published at NeurIPS 2023; a 'Cortex Architecture,' described as a parallel, distributed neural system for multi-modal learning; a 'Visual Cortex' module using binocular vision for 4D world reconstruction and semantic scene recognition; and memory-based Motion Control Models aimed at task-oriented, embodied-AI learning for robot control — consistent with press coverage describing the company's approach as a 'brain plus cerebellum' architecture for robot memory and control. The company also runs data-collection pipelines including intelligent teleoperation, first-person data capture, a CAD-to-multimodal-LLM (CAD-MLLM) tool, and 4D scene reconstruction tooling to support training its models. According to job postings, the company operates an office in Shanghai and is actively hiring research-engineer interns in areas including multimodal research and 3D research, alongside HR functions, suggesting a small, early-stage team. Company registration references also point to a Guangdong registration in addition to Shanghai operations. Public reporting indicates the company raised an angel funding round around July 2026, with founders and early team members reported to be affiliated with an HKU professor, though the company's own website does not name individual founders. The company's website footer copyright notice reads '2024-2025 Engram,' consistent with a company origin around 2024. No employee headcount, detailed funding amount, or named commercial customers have been published on the company's own site as of this writing.