Document AI — Épure 3D

Document AI is an evidence-first learning workspace for scanned textbooks and descriptive geometry.

It keeps the original 2D source drawing visible while providing an interactive Épure workspace for mathematically defensible 3D reconstruction. Exact figures can be rotated, measured, and inspected through Planche, Front, Top, and Isometric views. When the source drawing does not determine a missing coordinate, the application shows the complete red locus instead of inventing depth. Students can place an explicit red hypothesis, which remains clearly labelled as a user assumption.

OpenAI visual reasoning helps inspect figures, recover trapped text, create grounded explanations, and critique visual results. The 3D reconstruction itself is deterministic and testable: it is based on authored projection readings and closed-form descriptive geometry rather than automatic guessing.

The project was built with Codex and GPT-5.6, including the React and Three.js workspace redesign, geometry reconstruction and diagnostics, OpenAI Responses API integration, source/3D synchronization, red-point completion, automated tests, and public release.

Repository: https://github.com/PascalBurume/document-ai-v2

Built With

• codex
• express.js
• gpt-5.6
• openai
• react
• sqlite
• three.js
• typescript
• vite