Concept model
Assigning machining features to each face

Face-level predictions are grouped into pockets, counterbores, holes, steps and chamfers. The rounded outer corners are not fillets; they belong to the outer profile together with the side walls.

Actual project for this workflow
  1. 1Pocket9 faces
  2. 2Counterbore2 places
  3. 3Through-hole4 places
  4. 4Step (offset/ledge)2 faces
  5. 5Outer profile8 side faces
  6. 6ChamferC2 · 6 faces
Drawing
A-01
Name
Pocket/stepped block
Step
02 Feature recognition
Unit
mm
Author
Gyu-tae Kim
Updated
Product currently being developed with the team — in progressWedraw Pro

A product system where design data flows seamlessly into BOMs, technical drawings, and machining files.

  1. Furniture Builder Design
  2. Drawings and procurement
  3. Connector output
  4. Entering production

Two research projects on AI that recognizes 3D shapes, and a team product that turns furniture designs into factory data.

  1. Product View01 · AI·CAD/CAM·BlueChain add-inAI that reads 3D models and designs processing strategiesWe established an end-to-end workflow—from recognizing machining features to generating Fusion CAM processes and verifying the final results.ResponsibilityWe developed the geometry analysis, machining strategy, and Fusion integration components; the digital twin component was used solely for visualizing the results.9 typesMachining feature to identify · Include stock
  2. Research render02 · AI · Furniture · Parametric DesignAI that recognizes furniture panels and calculates resizing rulesPanels are identified and size adjustments are calculated using parametric formulas, enabling real-time design and ordering via the web.ResponsibilityWithin the team, I was responsible for AI-based panel recognition and the formulation of mathematical equations to calculate changes in panel size and position.15 typesPanel types recognized by AI
  3. Product View03 · Product · SketchUp · Furniture ManufacturingWedraw Pro, from furniture design to productionThe SketchUp design was output as BOM, label, drawing, and processing code and checked on the web.ResponsibilityWithin the team, I led the output of manufacturing data and production integration, along with the initial development of technical drawings and procurement processes.4 typesBOM, label, processing drawing, processing code output
02

Project Archive

Development work I took on beyond the key projects.

OpenCAD: free online CAD and 3D viewers

All tools and usage instructions.

I build a 2D CAD for opening, editing and saving DWG/DXF drawings, and viewers for STEP/SketchUp models and NC toolpaths. All are free personal open-source projects that run without installation, and the 3D core the viewers share is public too.

  1. 2D drawingOpenCAD DWG·DXFA free online CAD editor for opening, drawing, editing and saving DWG and DXF files without installation or sign-up. Measure distances, angles and areas, and export PDF, PNG or SVG..dwg.dxf
  2. 3D CAD parts/assembly modelsOpenCAD STEP ViewerThis is a free viewer that opens STEP·STP 3D CAD files in the browser without installation. You can check the assembly structure and part properties, measure hole diameters and distances between faces, and look inside with section and exploded views..step.stp.p21
  3. SketchUp 3D modelOpenCAD SketchUp ViewerThis is a free viewer that opens SKP files saved in SketchUp 6-2026 in your browser. You can look through tags, materials, groups and components, tour the space with scene playback and walk mode, and measure distances and angles..skp
  4. G-code·Heidenhain·MPR·PGI·XMLOpenCAD NC ViewerA free NC viewer for checking 3D toolpaths in G-code, Heidenhain (Klartext and DIN/ISO) and panel machining files (woodWOP MPR, WangShi PGI, KDT XML). You can trace code lines together with their moves and replay the machining sequence while watching the stock being cut. Lines to check before the program goes to the machine are flagged as warnings..nc.gcode.ngc.h.mpr.pgi.xml
WebGL2 library for developersOpenCAD 3D Viewer CoreThis is a WebGL2-based 3D viewer library used by NC, STEP, and SKP viewers. It provides common functions to receive, display, select, and measure scenes created by an interpreter.

Manufacturing Software Development Notes

All notes

Notes on what I worked through while developing AI, CAD/CAM and furniture manufacturing software. My role and results in each project are in the cases linked from each note.

03

AI research conducted through hands-on implementation

View full study history

As an undergraduate researcher, I worked on reinforcement learning, clothing classification, and virtual try-on. The images below were made by running the public code.

  1. Reinforcement Learning · Graph Neural Networks · Route search exampleA graph model for generating TSP routes using reinforcement learningI implemented a graph-based TSP search program that uses reinforcement learning to select the next city to insert into the route.ResponsibilityI connected the graph environment and training loop with the rules for inserting cities into the route.Step 2AI city selection + heuristic route insertion
  2. Computer vision · Image classification · Fashion-MNIST prediction exampleA model for classifying clothing photos with ResNetI trained a model to classify K-Fashion photos as dresses, outerwear, trousers or tops and analyzed errors for each class.ResponsibilityImplemented transfer learning, class-based sampling, and confusion matrix evaluation.4 typesClassification: Dresses, outerwear, pants, and tops
  3. Computer vision · Virtual try-on · Example of preprocessing for pose/human body parsingVirtual Try-On App for Clothing on Photos of PeopleI built the Android-to-server workflow that returns a virtual try-on image after the user selects a person photo and a garment.ResponsibilityThe input preprocessing pipeline for the public model has been integrated with a Flask API and an Android application.Step 3Pre-processing → Generate "try-on" image → Return to app
04

Patents and software programs

  1. PatentApplication · Co-inventionApparatus, method, and recording medium for 3D furniture modeling using the analysis of topological connections between furniture componentsAI Research and Development Manager / Co-Inventor ·Application Date 2025.11.21 · Application Number 10-2025-0177712
  2. ProgramRegistration application receivedLabeling Software for Training AI to Recognize Machined Shapes and FeaturesOverall Program R&D ·Creation Date 2025.05.13 · Date of registration application receipt 2025.12.12
05

Introduction

Profile photo of Sean Kim

I want to make people's work and daily lives a little easier with AI that can be applied in the real world.

In the research phase, we implemented methods to analyze geometry and learn design rules. In the product phase, these were applied to design modifications, the generation of manufacturing data, and pre-production verification. During operation, we collected processing results and error logs to identify overlooked issues and address their root causes. We hope that tools like these will reduce the burden of re-entering design information and minimize trial-and-error during the machining process, enabling even small-scale manufacturing workshops to leverage accumulated technical expertise.

AI
PythonTensorFlowPyTorchGraph neural networkReinforcement LearningComputer vision
3D · CAD/CAM
OpenCascadeSTEPAutodesk Fusion APICAM
Product Development
SketchUp Ruby APIJavaScriptFlask
Inquiries regarding recruitment or manufacturing technology collaboration