# Koji Itano | AI Robotics Engineer

> Personal portfolio and technical blog by Koji Itano (板野光司), an AI Robotics Engineer specializing in industrial automation, computer vision, and intelligent systems. Founder & CEO of INDUSTRIATE Inc. (インダストリエイト株式会社). 10+ years spanning Toyota Production System and Silicon Valley AI startups.

## Main Pages

- [Home](https://www.kojiitano.com/): Portfolio overview, skills summary, and featured projects
- [Experience](https://www.kojiitano.com/experience): Career history — Toyota, Connected Robotics, OSARO, RIOS, Enerza, INDUSTRIATE
- [Skills](https://www.kojiitano.com/skills): Technical skills matrix — AI/ML, robotics, computer vision, edge computing
- [Expertise](https://www.kojiitano.com/expertise): Domain expertise in manufacturing AI and industrial automation
- [Blog](https://www.kojiitano.com/blog): Technical articles on manufacturing AI, computer vision, RAG, and robotics
- [Contact](https://www.kojiitano.com/contact): Contact form

## Blog Posts — Manufacturing AI & Computer Vision

- [New Year 2026: My Resolution to 'Implement Intelligence in Industry'](https://www.kojiitano.com/blog/new-year-2026-manufacturing-ai-vision): After 5 years in Silicon Valley, returning to Japan and founding INDUSTRIATE to bring AI to manufacturing
- [Feeding Office Files to AI: Excel, Word, and PowerPoint Processing for RAG (2025)](https://www.kojiitano.com/blog/office-files-for-ai-rag-2025): How to transform Office documents into AI-digestible formats using MarkItDown, Docling, and VLMs for enterprise RAG
- [Local AI Hardware Selection Guide (2025)](https://www.kojiitano.com/blog/local-ai-hardware-selection-2025): Apple Silicon vs NVIDIA for inference and training — benchmarks, cost analysis, practical recommendations
- [From Physical Books to AI Knowledge Base (2025)](https://www.kojiitano.com/blog/books-to-ai-knowledge-base-2025): Non-destructive scanning, VLM-based OCR, and PDF-to-Markdown pipeline for RAG
- [Agentic Web (2025): From Search to Autonomous Action](https://www.kojiitano.com/blog/agentic-web-evolution-2025): The Web evolving from human search to autonomous AI agents — Intelligence, Interaction, and Economy dimensions
- [TubeletGraph: Tracking Object Transformations with Foundation Models](https://www.kojiitano.com/blog/tubelet-graph-object-transformation-tracking): AI tracking objects through transformations using SAM 2, CLIP, and GPT-4 (NeurIPS 2025)
- [From RNN to Transformer: The 30-Year Evolution of Sequence Models](https://www.kojiitano.com/blog/rnn-to-transformer-evolution): LSTM, GRU, Attention, and Transformer architectures explained from CS fundamentals
- [RAG in Manufacturing Part 1: Vector and Graph Fundamentals](https://www.kojiitano.com/blog/rag-vector-and-graph-fundamentals): Chunking, Vector RAG, and GraphRAG for reducing hallucination in factory knowledge
- [RAG in Manufacturing Part 2: Agentic Workflows and Prescriptive PARAM](https://www.kojiitano.com/blog/rag-agentic-and-prescriptive-param): Plan-reflect loops, tool use, and PARAM-style prescriptive workflows for MTTR reduction
- [Image AI Genealogy (2012-2025): From ResNet to CLIP and SAM](https://www.kojiitano.com/blog/image-ai-genealogy-2012-2025): Map of backbones, detection heads, self-supervised learning, and multimodal milestones
- [The YOLO Line (2016-2025): Speed, Tricks, and New Theory](https://www.kojiitano.com/blog/yolo-evolution-2016-2025): From v1 real-time breakthrough to v13 hyper-correlation — what changed and how to pick
- [Reinforcement Learning Genealogy (1992-2025)](https://www.kojiitano.com/blog/reinforcement-learning-genealogy-1992-2025): Q-learning to RLHF/DPO — milestones, why each mattered, and when to use them
- [Factory Knowledge RAG with Local LLMs](https://www.kojiitano.com/blog/factory-knowledge-rag-local-llm): Privacy-preserving RAG from SOPs and drawings with citations, local LLM orchestration
- [Technical Drawing AI: Multimodal Architecture and Edge Deployment](https://www.kojiitano.com/blog/technical-drawing-ai-guide): Vision transformer + layout-aware text model, cross-attention fusion, Jetson/NPU optimization
- [Industrial Camera Selection Guide](https://www.kojiitano.com/blog/industrial-camera-selection-guide): GigE/2.5G/5G/10G bandwidth, IP67, HDR, and line-scan practical selection matrix
- [Choosing Visual Inspection Solutions: Vendor Archetypes](https://www.kojiitano.com/blog/visual-inspection-vendor-archetypes-guide): Consulting, subscription, and SI archetypes matched to budget, speed, and customization
- [Creating Defects in Blender: Synthetic Data for Manufacturing AI](https://www.kojiitano.com/blog/blender-synthetic-defect-data): Solving data starvation with procedural defect generation — scratches, dents, cracks, warping
- [40 Years of 2.5D Visual Inspection](https://www.kojiitano.com/blog/2-5d-visual-inspection-history): From 1980s photometric stereo to event cameras and neural rendering
- [DINOv3 and Foundation Models for Industrial Inspection](https://www.kojiitano.com/blog/multimodal-llms-factory-surveillance): DINOv2, DINOv3, vision foundation models for manufacturing quality control (80+ papers)
- [Why I Started This Blog](https://www.kojiitano.com/blog/welcome-to-my-blog): Bridging research papers and production systems in AI robotics and manufacturing