NetsPresso On-Device AI Model Optimization Platform

Nota AI (Nota Inc.) is a South Korean on-device AI company headquartered in Daejeon, with offices in Seoul, Sunnyvale and Berlin. After starting with an AI product for reducing typing errors, it pivoted to on-device AI and now helps chipmakers, device makers and enterprises fit large AI models onto constrained hardware without losing accuracy.

NetsPresso takes a trained model and compresses, quantizes and benchmarks it for a specific target such as Qualcomm, Renesas, Arm, NVIDIA or Samsung silicon, producing a deployable model. Users can run the pipeline through tools and AI agents, including an agentic optimization loop for edge and server environments, or engage Nota's experts. The platform is integrated with NVIDIA TAO and Qualcomm AI Hub, and Nota also builds on-device vision solutions for intelligent transport, industrial safety, surveillance and driver monitoring. Customers and partners include Samsung Electronics, Arm, Renesas, Qualcomm, LG Electronics, Deutsche Telekom and FuriosaAI.

• Model compression: pruning and optimization that keep accuracy while cutting model size.
• Quantization: converts models to low-bit formats suited to NPUs, MCUs and GPUs.
• On-device benchmarking: measures latency and memory on target hardware such as Qualcomm and Renesas chips.
• Agentic optimization: AI agents automate the optimization pipeline for edge and server targets.
• Ecosystem integration: works with NVIDIA TAO, Qualcomm AI Hub and Arm tooling.
• Vertical solutions: on-device vision for ITS, industrial safety, surveillance and driver monitoring.

A Vietnamese camera or IoT device maker can optimize detection models to run on low-cost Arm or Renesas chips instead of adding a GPU. A smart traffic integrator can deploy optimized vehicle and plate recognition models on roadside edge devices. An enterprise AI team can quantize a local language model to cut GPU memory before rolling it out on-premise.

Shrinks models so they fit on cheaper chips: the company cites 70% memory reduction, 125 times faster inference on an MCU for a device maker and a 50% GPU reduction for a MoE language model. Hardware-aware benchmarking shows real latency on the target device before deployment. Teams can choose self-service agentic tools or expert services. Nota reports that its video AI cut surveillance infrastructure costs by up to 85% at more than 20 enterprises.

Phones, PCs, cars, robots and cameras are shipping with NPUs, but most AI models are too large to run on them without optimization. Chipmakers increasingly bundle optimization tools with their silicon, and enterprises in Asia want video analytics and language models on local hardware, creating demand for hardware-aware compression that works across many chip vendors.

Headquartered in Daejeon, South Korea, with offices in Seoul, Sunnyvale (Nota America Inc.) and Berlin (Nota AI GmbH). Raised seed, pre-Series A, Series A (6.6 million USD), Series B (14.6 million USD) and Series C (19.9 million USD) rounds. Holds a multi-year contract with Samsung Electronics for mobile, automotive and AI PC chips, and is a core technology partner in Korea's AI-specialized pilot city project. Shows an LRQA certification badge.