Technical Architecture — Index¶
Last synthesized: April 2026
This section documents the technical foundations of RapidDraft's CAD integration, geometry reasoning, error detection, and manufacturing intelligence systems.
Documents¶
| Document | Purpose | Audience |
|---|---|---|
| Drawing_Error_Detection.md | Taxonomy of drawing and CAD errors RapidDraft detects; five-pack check system; detection mechanism | CAD engineers, QA leads, product managers |
| NX_Graphy_Knowledge_Graph.md | Knowledge graph representation of CAD parts; extraction via NXOpen; application to parametric dimensioning | ML engineers, CAD integration engineers, geometry specialists |
| Vision_Model_Integration.md | Role of vision models in drawing validation; what they do well vs. limitations; confidence handling | Product managers, QA, vision/ML team |
| CAD_Viewer_and_Annotation_Architecture.md | Viewer stack choices, B-Rep vs mesh trade-offs, and the annotation-layer architecture for 3D and drawing review | Frontend engineers, CAD platform engineers, product leads |
| CAD_Drive_Collaboration_Mode.md | Product and architecture plan for the Google Drive / Docs-like CAD collaboration workspace: sharing, pinned comments, versions, non-STEP imports, and eval-driven rollout | Product leads, backend engineers, frontend engineers |
| Collaboration_Infrastructure.md | Production collaboration architecture: model-attached comments, auth, users, permissions, share links, screenshots, activity recovery, conversion, and small-team concurrency | Product leads, backend engineers, platform engineers |
| Database_and_Related_Services.md | Railway Postgres, Stytch auth, share links, artifact storage, backup, and embedding-service boundaries for the collaboration rollout | Product leads, backend engineers, deployment agents |
| Codex_App_Server_and_Programmable_CAD_Bridge.md | Explains how RapidDraft can embed real Codex, use ChatGPT sign-in, and control native CAD sessions through a programmable local bridge for NX first and HyperMesh later | Product leads, CAD platform engineers, AI/automation engineers |
| NX Live Control Architecture | Explains how RapidDraft controls the real NX session for review drawings and why that layer sits beside the neutral geometry stack | CAD platform engineers, product leads, automation engineers |
| Runtime_Architecture_Direction.md | Captures the current runtime direction: preview-first rendering, staged canonical detail, and decoupling exact extraction from baseline responsiveness | Product leads, CAD platform engineers, backend engineers |
| Drawing_Analysis/_INDEX.md | Overview of the Label Studio training track that supports RapidDraft drawing review | ML engineers, annotators, product leads |
| Drawing_Analysis/Label_Studio_Schema.md | Operational object classes, error labels, metadata, and pipeline rules for v1 labeling | ML engineers, annotators |
| Drawing_Analysis/Visual_Label_Reference.md | Inline visual examples that keep annotation regions consistent across collaborators | Annotators, reviewers |
| Code_Review/_INDEX.md | April 2026 code review — prioritized findings for rapiddraft_utumpitch, formatted for consumption by Codex |
Engineering leads, backend/frontend maintainers |
| Code_Review/09_Runtime_Pipeline_Audit_Synthesis.md | Places the inbox runtime audit into the code-review system and highlights the highest-ROI runtime changes | Engineering leads, backend/frontend maintainers |
Core Concepts¶
Knowledge Representation¶
RapidDraft extracts CAD models into a typed knowledge graph: Part → Features → Bodies → Faces → Expressions. This schema enables algorithmic reasoning about design intent, manufacturing constraints, and automatic dimensioning.
Error Detection (Packs A–E)¶
- Pack A: Drawing spec and compliance (format, standards, metadata)
- Pack B: Dimensional completeness (all features dimensioned, tolerances specified)
- Pack C: Tolerance and GD&T logic (datum priority, symbol validity, stack-up)
- Pack D: Manufacturing feasibility (DFM, tool access, thin walls, surface finish)
- Pack E: Assembly and interface logic (mating, clearance, fastener specs)
Vision Model Scope¶
Vision models extract drawing callouts and flag specification–geometry mismatches. Not used for 3D manufacturability checks (use CAD geometry for that). Emphasis on explainability and engineer-in-the-loop validation.
Integration Points¶
- NX Integration: NXOpen API for topology extraction, feature iteration, expression management, and in-session live control for review drawing workflows
- Codex Integration: local Codex harness for thread state, approvals, skills, and programmable bridge access into native CAD sessions
- Neutral Format Pipeline: STEP/IGES fallback when native CAD access unavailable
- NX Live Control: in-session host used when the workflow must stay inside the real NX drafting session rather than a detached helper or a neutral-file export path
- Drawing Memory: Captured check outcomes re-applied across CAD revisions
- AI Dimensioning: KG informs where vision and heuristics suggest dimensions
- Manufacturing Check Engine: Deterministic rules on CAD B-Rep (Packs B–E); vision for drawing validation (Pack A)
Technical Dependencies¶
- NXOpen: Siemens NX geometry and feature APIs
- STEP/IGES Parsers: Neutral CAD format processing
- Vision Models: OCR and drawing specification extraction (local or cloud)
- Graph Storage: JSON (lightweight) or Neo4j (enterprise scaling)
- Rule Engine: Deterministic checks on geometry; ML-assisted confidence scoring
Roadmap Notes¶
- Phase 1 (MVP): Single-body part reasoning; NX + STEP support; basic feature recognition
- Phase 2: Multi-body assembly reasoning; Teamcenter integration; advanced GD&T logic
- Phase 3: Cross-supplier assembly tolerance analysis; real-time collaboration graph updates