Project Neev · v3 · AI construction platform
An AI construction pipeline for India. Neev reads the plot from any site photo, checks it against local bylaws, evolves legal floor plans, prices them against CPWD rates, and exports the drawings a contractor can actually build from.
The pipeline
Neev is a single connected pipeline, from the raw site photo to the export a contractor will open. No file shuffling between vendors — every stage hands clean data to the next.
Computer vision extracts plot dimensions and obstructions from a site photo.
modules/scan
FSI, setbacks and ground coverage checked against the city's bylaws before a wall is drawn.
modules/compliance
A genetic algorithm evolves 3–5 layout alternatives scored on efficiency and circulation.
modules/design
ML predictions calibrated against a deterministic CPWD DSR 2025 BOQ, with confidence bounds.
modules/estimation
Rooms extrude into textured 3D — viewable live in the browser via Three.js.
generate-3d
Material engine matches brands to budget tier with a five-factor sustainability score.
material_engine
DXF CAD with layers, PDF reports and standalone 3D HTML viewers generated on demand.
export-dxf · pdf
Dockerized backend and frontend behind Nginx, health-checked and ready to scale.
docker-compose
Product
Every system here solves one real headache on an Indian site. Three systems carry the pipeline; the rest of the stack picks up the paper around them.
01 · Design
Constraint-satisfaction across every room — adjacency, daylight, circulation and bathroom placement — iterated over hundreds of generations. Neev doesn't stitch rooms together from canned blocks; it breeds a plan to fit your plot.
02 · Cost
A Random Forest trained on CPWD DSR 2025-aligned data is blended against a deterministic government-published BOQ. When the ML model and the DSR disagree, the estimate shows you the gap — and the confidence band around it.
03 · Deliver
The 3D model is extruded straight off the generated plan — no manual rebuild — and exports are produced with the same geometry. The DXF opens in AutoCAD with proper layers; the 3D viewer runs from a single HTML file on any computer.
Datasheet
The support systems that make the three above hold up in production — each one small, typed, and unit-tested.
scan_routes
states_data.json
material_engine.py
update_prices.py
docker-compose.yml
/api/v1.app/api/v1
Workflow
Or type dimensions straight in. Nothing to install; Neev runs in the browser-facing Streamlit app.
City, bedrooms, floors, and room preferences — pooja, study, servant quarter, kitchen type, bathrooms.
Design, compliance, pricing and 3D all fire in one call. A few minutes of compute for a full package.
Plans ranked on efficiency, circulation and external-wall ratio, with cost broken down per alternative.
Grab the DXF for a draughtsman, the PDF for the client, and the 3D HTML for the site meeting.
Pricing data
Every estimate is anchored to each state's PWD Schedule of Rates, relative to a Bangalore baseline of 1.00 — CPWD DSR 2025 for Delhi. No national "one price fits all".
…plus every remaining district in India, banded by state PWD SOR calibration.
Bangalore 1.00Delhi 1.15Mumbai 1.22
For developers
Pydantic-validated requests, synchronous JSON responses, and a pipeline that fits into any estimator or floor-plan product. The whole thing runs on one machine if you want it to.
# five alternatives, priced & ready
curl -s localhost:8000/api/v1/generate-designs \
-H "Content-Type: application/json" \
-d '{
"city": "Bangalore",
"plot_area": 2000,
"num_bedrooms": 3,
"num_floors": 2,
"room_preferences": {
"has_pooja_room": true,
"has_study_room": true,
"kitchen_type": "closed"
}
}'
→ {
"designs": [
{
"alt": "A",
"efficiency": 0.71,
"gfa_sqft": 1995,
"estimated_cost": { "low": "43.2L",
"mid": "46.8L", "high": "50.1L" }
} …
]
}Endpoints — /api/v1
POSTcomplete-analysisphoto in → full package outPOSTgenerate-designs3–5 layout alternativesPOSTpredict-costsML + CPWD blendPOSTgenerate-3dextruded viewer + previewPOSTrecommend-materialsfive-factor scoringPOSTexport-dxf / export-pdfCAD + reportGEThealthliveness probeOpenAPI docs served at /docs in the running container.
Roadmap
The chart above is ground floor, not the ceiling. What's live today, and what's on the drawing board.
Project Neev
Run the live demo and watch a plot become a build. Or read the source — it's all there.