IN DEVELOPMENT -- NOT FOR PRODUCTION

Engineering drawings in.
Structured BOMs out.

VeXtraBOM reads drawing packages -- title blocks, parts lists, and tabulated variant tables -- and builds the bills of materials and configurator matrices your engineers assemble by hand today. Self-hosted on your own hardware, so your drawings never leave your building.

Build status // the extraction, roll-up, and workbook-generation prototypes run today against real manufacturing data. Packaging them into a product is in progress -- capabilities below are labeled honestly.
SEC 01

Capabilities

Domain logic for configure-to-order manufacturing -- not generic document extraction. Each item is marked by its current state.

Parts-list extraction Prototype

Reads item, part number, and quantity tables straight off a drawing -- including skewed phone photos -- and reconstructs them as structured rows.

Drawing-type identifier Prototype

Tell the console your sheet's layout -- parts-list, config/variant (BOM chart top-right, calculations below), or CAD PDF -- or let auto-detect find tables wherever they sit. On variant sheets, highlighter-marked rows are flagged as the selected configuration.

Quantity-chart capture Prototype

Finds the quantity and variant charts printed under the drawing -- the calculation tables -- and transcribes them verbatim into the draft workbook's QTY TABLES sheet. Cells that need a team decision are color-flagged, never guessed.

BOM data QC Prototype

Cross-validates BOM sheets against variant matrices and flags typos, orphaned tables, and broken references -- the errors that silently corrupt downstream configurators.

Super-BOM roll-up Prototype

Resolves the assembly structure across a whole drawing package into an indented multi-level BOM plus a consolidated purchase list. Numeric quantities multiply down the tree; variable quantities are carried unresolved and marked.

Vector-PDF fast path Prototype

CAD-exported PDFs carry exact text with coordinates -- no OCR involved, near-error-free extraction for modern drawing sets. Scanned pages fall back to the OCR engine automatically.

Workbook generation Prototype

A whole drawing package becomes a draft BOM workbook: parts sheets from the assembly structure, the drawings' quantity charts transcribed verbatim into a QTY TABLES sheet, and color-coded flags for the naming and calculation decisions that stay with your team. Highlighter-marked chart rows resolve their quantity codes into a SELECTED column beside the printed QTY (see SEC 02). Nothing is guessed.

SEC 02

It reads the ink

On a shop floor, the chosen configuration lives in a highlighter stroke -- digital on the PDF, or marker on a printed sheet. VeXtraBOM keeps what commercial parsers throw away: the drawing's markup layer and the color of every line it reads.

Highlighter-selection capture Prototype

When an engineer marks the chosen row of a configuration chart, VeXtraBOM reads which row was chosen and resolves it: quantity codes like (XE) and quoted part-number variables like "ZD" take their values from the marked row and are listed in a SELECTED column beside the printed QTY in the generated workbook, with every choice and its source recorded on a CONFIG SELECTIONS sheet.

Digital highlights in CAD PDFs are read from the PDF's own highlight and ink annotations -- exact geometry intersected with the embedded text, fully deterministic. Physical marker on flattened or scanned sheets saved as PDF is separated by stroke-color math on the rendered page pixels -- no model guessing what "looks highlighted". A mark that could mean two rows is flagged for review, an unresolvable code stays blank and color-flagged, and every flagged row stays editable. Nothing is ever guessed.

Variant table -- row flagged from ink
WIDTH PART NUMBER QTY
18" VX-100310_Frame18 1 EA
24" VX-100311_Frame24 1 EA
36" VX-100312_Frame36 1 EA
SEC 03

Architecture

Not a wrapper around someone else's cloud model. VeXtraBOM is assembled from EdgeXene's own open-source engines. Each layer is inspectable, and none of them opens a third-party data path.

Local perception layer

Text recognition runs on XLiteOCR, EdgeXene's own Apache-2.0, CPU-only engine, on the same machine as the product. Its document-reading playbook -- line-level recognition, crop-and-deskew preprocessing, stroke-color math -- was proven in zkVIPER, our on-device ID-verification engine, before it was pointed at drawings. The extraction pipeline makes no cloud calls and no model-API calls.

In-memory processing

Uploaded drawings are processed in memory and are not written to disk; results and exports are assembled in your browser. What the service does keep -- and it is short -- is spelled out in the privacy policy.

Deterministic extraction

No LLM sits in the extraction path. Tables are found and rebuilt with geometric reconstruction and explicit rules, so each extracted value traces to a location on the sheet -- and a wrong read is visible, not hidden inside a plausible answer.

SEC 04

Why self-hosted

Customer drawings are crown-jewel IP. VeXtraBOM is built to run entirely inside your walls.

Your drawings stay home

Nothing is uploaded to anyone's cloud. For shops handling export-controlled technical data, that is not a preference -- it is an eligibility requirement.

No per-page metering

Cloud document APIs charge per drawing, forever. A self-hosted engine costs the same whether you process ten drawings or ten thousand.

Explainable, auditable output

Deterministic checks and a review worklist instead of a black box. A wrong extraction costs a correction during review -- an estimator approves everything before it ships.

SEC 05

Follow the build

VeXtraBOM is in active development. If drawing-to-BOM automation is a pain you have, we want to hear about your drawing sets.

contact@edgexene.io
TitleVeXtraBOM -- drawing-to-BOM engine Drawn byEdgeXene LLC RevDEV StatusIN DEVELOPMENT PrivacyPOLICY ©2026