1 · AI extraction
A vision model reads your drawing and returns only geometry: room polygons, labels, doors, windows, staircases, fixtures and the North arrow. It performs no Vastu reasoning at all.
Architecture
How the analysis is produced
VastuLens separates interpretation from judgement. AI reads and explains; a deterministic engine decides. That separation is what makes the score reproducible and defensible.
2 · Canonical floor plan model
Everything is normalised into a single versioned data structure with confidence values and a source flag on each element, so AI-detected and user-corrected data are always distinguishable.
3 · Human verification
You confirm room types and set true North. Low-confidence detections are flagged. Analysis is blocked until orientation is explicitly established — we never guess North.
4 · Geometry & zone engine
The building outline is divided into the 3×3 Vastu grid rotated to true North. Each room gets an area-weighted zone distribution and a dominant zone, plus centre (Brahmasthan) occupancy.
5 · Deterministic rule engine
A versioned rule set evaluates each subject against preferred, acceptable and avoid zones. Findings carry a rule id, evidence and point contribution. The weighted score is pure arithmetic.
6 · AI explanation layer
Only after the score exists does a language model write the narrative and answer report questions — strictly grounded in the computed findings. It cannot change the score.