Lumira Atelier
AI-Co-piloted optical design wizard
A 5-step browser agent for smartphone main / wide cameras — real production seed matching, Optiland ray tracing, MTF/RMS evidence, and a downloadable PDF report.
- Category
- Optics × AI
- Year
- 2026
- State
- Live in production
- Stack
- Next.js 16React 19FastAPIOptiland 0.6GPT-Image-2
The wizard currently handles only smartphone main / wide cameras. Across five steps it refines focal length, F-number, FOV, image height, element count, and physical constraints; the backend matches against a real production design library and runs Optiland 0.6 ray tracing, then reports paraxial data, MTF/RMS evidence, candidate tradeoffs, risks, and next optimization actions. The PDF (GPT-Image-2 cover, Claude Opus 4.7 executive summary) lands as a download. This launch build is not the final designer replacement, but it can already produce a credible first-pass main-camera draft.
Real ray tracing, not a mock. Optiland 0.6 + rayoptics on FastAPI; React Three Fiber + ECharts on the front-end. 143/143 backend pytests green; 4 scenarios × 4 PDF pages rendered cleanly across cairo / poppler / mupdf backends.
Live at lumirahq.com/agent — referenced as 'most technically dense agent demo' by every reviewer who has run it through.
Working on something similar?
Start with a 30-minute call. We pick one project together that fits a pilot.