Why template OCR fails on drawings
Classic document extraction is template work. You draw zones on a reference page, the software reads the same coordinates on every copy, and the approach holds as long as each page looks like the reference.
Drawings break that assumption immediately. Every originator designs its own title block. One archive mixes A4 permits with A0 site plans, horizontal notes with vertical dimension strings, printed tables with stamps, revision clouds and handwritten markups. A template tuned to one firm's border fails on the next firm's, and scanned legacy drawings add skew and noise on top.
MYX Analyzer does not use zone templates. It reads each sheet as a whole page and answers questions you write in plain language, one prompt per field. The same field definitions carry across originators, disciplines and decades of drafting styles, because they describe what you want, not where it sits on the page. Setup becomes a list of questions rather than a labeling project: write the prompt for a field once and apply it to the whole archive.
85% of enterprise data is unstructured. Drawing archives are a textbook example: dense, visual, and invisible to systems that expect neat text.
Title blocks, revisions and site plans as structured fields
In MYX Analyzer, a drawing register is a table. Each document is a row, each fact you care about is a column, and each column carries a short plain-language prompt that tells the Analyzer what to extract. Technical Docs & Drawings is one of the platform's published use cases, alongside several that lean on drawings just as hard:
The field list is yours to define. A typical register starts with:
| Field | What the prompt asks for | Where it typically sits |
|---|---|---|
| Drawing number | The document identifier, exactly as printed | Title block |
| Revision | The latest revision letter and its date | Revision table |
| Originator | The firm that produced the sheet | Title block |
| Title and discipline | What the sheet shows and which trade it belongs to | Title block |
| Scale and sheet size | The stated scale and paper format | Title block |
| Equipment shown | Structures and hardware visible in the plan or elevation | Drawing area |
Prompts live at field level, and a per-table project context describes the set as a whole, so the Analyzer reads each sheet knowing whether the table holds tower structurals or substation layouts.
Tower portfolios push this further: site plans and structural sheets sit in the same shared workspace as leases and permits, so the drawing register lines up with the rest of the site record. The telecom towers page shows that portfolio view. When the site record is decades old and only half-trusted, a legacy site audit rebuilds it from the documents themselves.
High-detail media analysis
The hard part of a drawing is small. Dimension strings, cable schedules, note columns and revision dates are printed at sizes that survive an A0 plot but blur when the sheet is scanned, shrunk and passed through a generic pipeline. The information sits exactly where processing quality drops first: hairline leaders, small fonts, dense schedule tables.
MYX Analyzer includes media resolution settings for high-detail documents and images. For a drawing set you raise the resolution, so small print reaches the model legibly instead of as noise.
Reading everything does not mean trusting everything. Every extracted value returns a confidence indicator, HIGH, MEDIUM or LOW, so a date read from a faint fourth-generation photocopy arrives flagged for review rather than silently accepted.
From drawing sets to a searchable register
Getting an archive in is bulk work. MYX Analyzer imports from SharePoint and from bulk folders, and each imported document becomes a row in the grid.
Extraction then runs field by field across the whole set. Process hundreds of documents overnight with enterprise batch processing; before anything runs at scale, batch submissions show an upfront cost estimate. Runs are monitored in real time, and results stream into the grid field by field as they complete.
Review happens in a spreadsheet-style grid: rows are documents, columns are extracted fields. When a prompt needs refinement, you rerun a single cell or row instead of the whole archive. Shared workspaces with permissions keep the review distributed: a document controller, a structural engineer and a lease analyst can work the same register.
Each value keeps a reference to the page and section of the source document it came from. That reference is what turns a spreadsheet of extractions into an auditable register: checking a claim means opening the cited sheet, not leafing through the set.
When the register is ready, export it to Excel with custom templates, including an AI template designer that builds the export layout for you. During acquisitions the same workflow produces the document inventory for due diligence, and the grid itself is the same one teams use for lease abstraction.
Formats and languages
MYX Analyzer accepts PDF, JPG and MP4 files: documents, images and video. In practice that means issued drawing sets as PDF, photographed or single-sheet scans as JPG, and site footage as MP4. Native CAD formats such as DWG are not on the list; plot them to PDF, the format sets are issued and archived in anyway, and upload those.
Language is not a filter either. The Analyzer supports 40+ languages, so an archive that mixes German structural sheets, French permits and English as-builts still lands in one schema.
FAQ
Can MYX Analyzer read scanned and legacy drawings?
Yes. The Analyzer reads the rendered sheet rather than relying on a clean text layer, and media resolution settings let you raise the detail level for dense or faint documents. Every value returns a HIGH, MEDIUM or LOW confidence indicator, so uncertain reads from older scans are flagged for review.
Does MYX Analyzer accept native CAD files such as DWG?
The Analyzer accepts PDF, JPG and MP4 uploads. Native CAD files are not accepted directly: plot or export the sheets to PDF, the format drawing sets are issued in, and upload the PDFs.
How do I verify that an extracted value is correct?
Every field carries a confidence indicator and a document reference to the page and section the value came from, so verification means opening the cited sheet. You can rerun any cell or row after refining its prompt, and extraction measures 94.7 % on internal evaluation sets.
Put your drawing archive in one grid
Upload a folder of sheets, define your fields in plain language, and review the first extractions before the 30-day trial ends.
30-day free trial · Card required — first charge only when the trial ends · Cancel anytime