NLPP
Fair-market price for any drawing.
Non-Linear Price Prediction. Three quantiles. Every feature explained in EUR.
bracket-v3.pdf
1.4 MB, paired with bracket-v3.jt
Inside NLPP
What NLPP delivers.
p10, p50, p90 in EUR.
Three quantile predictions from three LightGBM regressors. The spread tells you the certainty.
Every feature's contribution.
Exact tree-SHAP per-feature deltas in EUR. The sum equals the p50. No black box.
2 to 4 parts side by side.
Compare drawings on price, features, and cost breakdown. Spot the outlier.
Slide a feature. Re-predict instantly.
Adjust wall thickness, holes, volume. The model re-runs in milliseconds.
Where the overspend is.
Project-wide cluster analysis ranks parts by annual savings potential against your material master.
Try it with no upload.
Eight curated sample parts ship with the agent. Try it end-to-end with zero customer data.
How it works
From drawing to defensible price, in 4 steps.
UPLOAD
Open the drawing.
Pick a drawing already analyzed by the drawing-analysis worker, or upload a sample part from the gallery.
EXTRACT
Features pulled automatically.
13 features per part. Ten physics. Two market. One data-quality flag. Optional 3D geometry and material master.
Features extracted
PREDICT
Three quantiles in EUR.
p10, p50, p90 from three LightGBM regressors trained on 172 real parts. Returned in milliseconds.
EXPLAIN
See what drove the price.
Tree-SHAP per-feature deltas in EUR. Wall thickness, holes, material price. Every line traceable.
SHAP contributions
Connected
Plug NLPP into the data you already have.
Where the data comes from
Drawings come in via analysis. 3D and material master are optional Excel.
The cost agent shares parameter extraction so you don't pay twice. Material masters stay org-scoped, never shared cross-tenant.
Inside MAindTec
NLPP is part of the MAindTec family.
Lives next to Workspace projects. Shares parameter extraction with Cost. Org-scoped material masters never cross tenants.
Why NLPP
Built for fair-market pricing.
Without NLPP
Gut-feel pricing from what we paid last time
Negotiations without a number to defend
One supplier quote feels like the market
Overspend hidden across hundreds of parts
With NLPP
p10, p50, p90 from 172 real historical parts
SHAP per-feature deltas in EUR to show the supplier
Supplier and country offsets calibrated against the training set
Hot Spot dashboard ranks projects by annual savings potential
3
Quantiles (p10/p50/p90)
13
Features per part
8
Languages
172
Trained on real parts
Testimonials
Trusted by engineering teams.
The AI-powered analysis was professional, efficient, and faster than I expected.
Axel Neumann
Platform Product Manager, KUKA
FAQ