
Delivered local AI document processing for ERP-ready bills of materials, certificate extraction, and human review

Delivered Workflow
We delivered an on-premise document workflow that turns technical drawings and spreadsheets into bills of materials for human review. Local AI and vision run on NVIDIA DGX Spark alongside deterministic DXF and Excel parsing. Geometry analysis and OpenCV identify triangle marks associated with items requiring EN 10204 3.1 certificates, while OCR extracts supplier certificate data for Access. Semantic retrieval queries the ERP's SQL database to match extracted items with existing records. Reviewers check item matches and certificate requirements before approving Excel or CSV output in the ERP layout. Company data remains on site throughout processing.


Established the document review workflow
We mapped technical documents, material lists, and certificate requirements into a local processing workflow. Human review was built into the path to ERP import, with company data kept on site throughout document analysis, item matching, and export preparation.
Implemented parsing and local AI
We deployed local AI and vision on NVIDIA DGX Spark and implemented deterministic DXF and Excel parsing. Geometry analysis and OpenCV detect certificate triangle marks, while OCR extracts supplier document information for review and import into Access.


Delivered ERP-ready review and export
We connected semantic retrieval to the ERP's SQL database to match extracted items with existing records. Reviewers check these matches and certificate requirements, then approve Excel or CSV material lists structured for the ERP's required import layout.
Tested outputs against source documents
We compared extracted geometry, item matches, certificate marks, and exported rows with representative source documents. Testing covered the review path for ambiguous results and verified that approved material lists and certificate data reached their intended ERP and Access outputs.

