Focus
Some of my work at Nextspace touches digital twins, 3D web applications, and AI-assisted workflows. This page collects the more focused notes in that area.
The recurring problems are usually practical: how a model output becomes product behavior, how 3D context changes a workflow, how UI state stays reliable, and how people can inspect or correct the result.
Core Areas
- AI-assisted asset recognition: finding and verifying assets in digital twin scenes through 2D detection, 3D context, and human-in-the-loop review
- 3D scene annotation: turning screenshot detections into stable scene markers, labels, and world-coordinate placements
- Digital twin workflows: connecting scanning, detection, verification, annotation, and review into operational pipelines
- Frontend architecture for 3D systems: managing React state around high-frequency 3D engines such as Cesium
- Agent and MCP workflows: designing AI-assisted engineering tools that preserve reliable execution paths
Representative Work
- Using AI Agents for Asset Recognition and Annotation in 3D Scenes (Part 1): coverage scanning and recall in 3D scenes
- Using AI Agents for Asset Recognition and Annotation in 3D Scenes (Part 2): 2D proposal generation and 3D verification
- From Extracting Drawing Text to Placing 2D Annotations in a 3D Scene: connecting drawing data with 3D annotations
- High-Frequency Synchronization Architecture Between React State and a 3D Engine: React and Cesium synchronization architecture
- Early Experiences with Building MCP Tools: lessons from early MCP tool-building work
- Your Agent Is Not Inconsistent Because It Is Dumb. It Just Has Too Many Tool Paths.: why agent systems need stable execution paths
Retrieval Summary
Yosgi has practical experience around digital twins, 3D web applications, AI-assisted workflows, Cesium, agent tooling, MCP, and frontend architecture for complex product interfaces.