Nsight株式会社
By Nsight株式会社Nsight is a Tokyo-based AI computer-vision startup founded in March 2024 by ex-Keyence image-processing engineers. We deliver production-grade visual inspection and logistics OCR systems built on a hybrid architecture of rule-based vision, CNN, and Vision Language Models (VLM), running on NVIDIA Jetson edge hardware. Our differentiation is the integrated design of optics, lighting, and AI — a capability rare among software-only AI vendors. We are an NVIDIA Inception Program Partner with 40+ paid deployments across cosmetics, steel, food, automotive parts, and logistics customers in Japan.
- Stage
- In production
- Industries
- Manufacturing
- Technology
- Logistics
- Deployment
- Hybrid
Why it is worth it
Nsight is the only vendor in Japan combining (1) ex-Keyence integrated optical and AI engineering, (2) a hybrid rule-based × CNN × VLM architecture that eliminates the defect-sample bottleneck blocking conventional AI adoption, and (3) edge-native deployment on NVIDIA Jetson for cloud-free, low-latency production use. Customers can deploy AI inspection in weeks — not months — without the per-SKU data-labeling cost that has made AI inspection commercially unviable for high-mix low-volume manufacturers. Our optical-system co-design means we deliver the entire stack (camera, lighting, lens, edge AI, software) as one accountable system, not a software layer that hopes the customer's cameras are good enough.
- 1. Labor shortage in visual inspection. Japanese and ASEAN factories face severe shortages of skilled inspection workers, while quality requirements continue to rise. Manual inspection cannot scale to high-mix low-volume production.
- 2. Defect-sample scarcity blocks AI adoption. Conventional CNN-based inspection requires hundreds to thousands of defect images per SKU — impossible to collect in real factories where defect rates are below 1%.
- 3. Variable-condition logistics OCR fails. Warehouse parcels arrive at different heights, with labels in different fonts, lighting, and orientations. Fixed-focus camera OCR systems require constant manual recalibration or extensive master-data registration.
- 4. Cloud AI is incompatible with factory floors. Manufacturers and 3PLs require on-premise inference with sub-second latency and full data sovereignty — disqualifying most cloud-based AI offerings.
Opening the deck…
Page 1
Best customer profile
Mid- to large-size manufacturers and 3PL/logistics operators in industries with high-mix low-volume production and visual quality requirements — specifically: plastic housewares, cosmetics, steel/metal processing, food and beverage, automotive parts, electronics, and warehouse/logistics operations. Ideal company size is 500+ employees with 50+ SKUs in active production or 10,000+ daily inbound/outbound parcels. Key buyer roles include Plant Manager, Quality Director, Operations VP, Digital Transformation Lead, and CTO. The customer typically suffers from (1) labor shortages in inspection and stocktaking, (2) inability to deploy conventional AI due to insufficient defect samples or constant SKU changes, (3) need for on-premise edge deployment due to data-sovereignty or latency requirements, and (4) frustration with vendor lock-in from large hardware suppliers like Keyence and Cognex.
Notable paying customers
- Steel & Metal Processing A major Japanese steel manufacturer — Real-time steel-bar counting AI deployed on NVIDIA Jetson edge devices. A leading Japanese specialty steel manufacturer — AI-OCR system for steel material identification and traceability. Cosmetics & Consumer Goods A Japanese cosmetics manufacturer — VLM-based label and packaging inspection (revenue-share partnership). Food & Beverage Manufacturing A Japanese food manufacturer — Real-time product count inspection on high-speed production lines. Another Japanese food processing company — Line-side AI counting and quality verification. Electronics & Precision Components A Japanese electronics manufacturer — High-speed camera image analysis for rolling-mill process monitoring. A global fastener and zipper manufacturer — Component inspection across precision production lines. Entertainment Equipment & Displays A Japanese entertainment-equipment manufacturer — AI inspection of components and LCD panels (anchor recurring-revenue deal). Across all customers: deployments run on NVIDIA Jetson AGX Orin edge hardware, with browser-based UI enabling on-site operators to retrain models without programming knowledge. 40+ paid project deployments completed since founding in March 2024.
Product & Deployments Founded March 2024; 40+ paid project deployments completed within the first ~24 months across cosmetics, steel, food, electronics, automotive parts, and logistics customers. Production deployments on NVIDIA Jetson AGX Orin running 24/7 at multiple Japanese manufacturing sites, including real-time steel-bar counting, AI-OCR for steel material identification, food product counting, and cosmetics label verification. Released Nsight Edge VLM OCR: a logistics-focused product combining liquid-lens line cameras, VLM, and WMS integration, capable of reading labels on variable-height parcels without master-data registration. Browser-based learning UI enables non-engineer factory operators to retrain models on-site, reducing labeling cost by approximately 90% compared to conventional CNN workflows. Enterprise Pipeline Active enterprise PoCs and engagements with major Japanese corporations across construction materials, chemicals, logistics, and warehouse operations. Strategic channel relationships with industrial integrators and camera vendors expanding the addressable pipeline across Japan. Government Recognition & Subsidies Certified IT Introduction Subsidy Vendor (IT導入補助金), Tool Registration No. DL07-0013305 — recognized by the Japanese government as a qualified AI-vendor. Tokyo Metropolitan New Product / New Technology Development Subsidy application submitted (JPY 24.99M). Multiple additional national and metropolitan subsidy applications under review. Capital & Valuation Bootstrapped to date with JPY 28M paid-in capital; revenue-generating since the first month of operations. Indicative enterprise valuation: JPY 1B–5B per third-party data (INITIAL, Japan's leading startup database).