Computational Design Developer — Scripted Furniture Pipeline

We're an Australian custom furniture company that designs original pieces for hospitality venues. We generate photorealistic furniture concepts using AI image tools, and we manufacture at commercial scale — but right now there's a slow, manual gap between "approved concept image" and "drawings a factory can build from." We want to close that gap with a fully scripted pipeline, and we're looking for someone to architect and build it with us. The pipeline we want to build Stage 1 — Multi-view renders → 3D mesh. Take a set of consistent rendered views of a furniture piece (front, side, three-quarter, etc.) and produce a clean 3D mesh. We're open on approach: AI image-to-3D services via API (Tripo, Meshy, Hunyuan3D, TRELLIS, or similar), photogrammetry adapted for synthetic imagery, or a hybrid. You should know the current landscape well enough to recommend what's production-viable today, not just what demos well. Stage 2 — Mesh → manufacturable CAD. Raw AI meshes aren't buildable. This stage cleans, retopologises, and converts the mesh into a proper CAD representation — NURBS/BREP surfaces or a parametric reconstruction — with real-world dimensions, correct wall thicknesses, and geometry a factory can actually machine, cast, or fabricate. We currently work in Rhino 8, so Rhino.Compute / rhino3dm / Grasshopper experience is highly relevant, but we're open to your recommended stack if you can justify it. Stage 3 — CAD → shop drawings. Scripted generation of dimensioned 2D shop drawings from the CAD file: orthographic views, sections, key details, exploded views where relevant, title blocks, and a basic bill of materials. Output as PDF + DXF/DWG. The goal is that a human reviews and annotates, rather than draws from scratch. Stage 4 — Everything callable from the web. The whole pipeline needs to run headless, triggered by API calls, so it can sit behind an HTML-based product configurator. A user (or our team) adjusts a concept in the browser, the pipeline runs server-side, and manufacturable files come out the other end. You don't need to build the configurator front-end — but the pipeline must be architected as clean, documented services/endpoints that a front-end can plug into. What you'll deliver Architecture document — recommended tools, services, and data flow for the full pipeline, with honest notes on what's reliable now vs. what needs human-in-the-loop checkpoints. Working proof of concept — one real furniture piece (we'll supply the renders) taken end-to-end: renders in, dimensioned shop drawing set + BOM out, via script. Production pipeline — the PoC hardened into documented, repeatable code (Python preferred) with a defined API surface for the configurator integration. Handover — clear documentation and a walkthrough session so our team can run, maintain, and extend it. Required skills Deep experience with mesh processing and repair (retopology, decimation, watertight checking) — e.g. Blender scripting, MeshLab, Open3D, PyMeshLab Rhino 8 / Grasshopper / Rhino.Compute / rhino3dm, or an equivalent programmatic CAD stack (FreeCAD + Python, Fusion API, OpenCascade), with mesh-to-BREP conversion experience Automated 2D drawing generation — scripted views, sections, dimensioning, and layout export to PDF/DXF Strong Python; comfortable building headless services and REST APIs Familiarity with current AI image-to-3D tools and their APIs, including their real-world limitations Understanding of what makes geometry manufacturable (tolerances, wall thickness, joinery logic, material constraints) — furniture, joinery, product design, or industrial design background is a big plus. How we'll work We'll start with a paid discovery/architecture milestone before committing to the full build. We're a small, fast-moving company — you'll work directly with the founder, decisions get made quickly, and there's meaningful ongoing work if the pipeline succeeds (this is the manufacturing backbone of a larger product vision). To apply, please answer Describe a project where you converted mesh data into manufacturable CAD or drawings. What broke, and how did you fix it? Which image-to-3D approach would you reach for first with synthetic renders (not photos) as input, and why? What's your honest view on how much of Stage 2 (mesh → clean CAD) can be automated today vs. requiring a human checkpoint? What's your preferred stack for Stage 3 (automated shop drawings), and can you share an example output? Applications that answer these specifically will be prioritised over generic proposals.