AI for Design Quality & DFM
Can Claude AI review engineering drawings with GD&T, tolerances, and BOM tables? We tested it. Here's what works and what doesn't.
·
⏱
8 min
Michelle Ben-David
Michelle Ben-David is a mechanical engineer and Technion graduate. She served in an IDF elite technology and intelligence unit, where she developed multidisciplinary systems integrating mechanics, electronics, and advanced algorithms. Her engineering background spans robotics, medical devices, and automotive systems.
BOTTOM LINE
Claude AI can't read engineering drawings in any meaningful way. It can look at screenshots, but it can't parse DXF/DWG files, reliably interpret GD&T, extract BOM data, or cross-reference drawings against 3D models and PDM systems. For real drawing review, you need AI that reads CAD natively and understands engineering data at a structural level. Leo AI does exactly that, backed by 3 US patents for CAD geometry reading and training on over one million vetted engineering sources.
Engineering drawings are dense. A single drawing sheet packs GD&T callouts, tolerances, revision blocks, title block data, BOM tables, section views, detail views, and notes into a format that's been standardized for decades. Reviewing them properly takes experience and attention to detail. It's exactly the kind of tedious but critical task that engineers wish AI could help with.
So can Claude AI review engineering drawings? The short answer: not really. Not in the way engineers actually need.
Claude is a powerful general AI model. It's great with text, solid at reasoning, and can discuss engineering concepts all day. But engineering drawings aren't just text. They're a specific format that combines 2D geometry, symbology, structured data, and spatial relationships in ways that general AI simply wasn't built to parse.
Let's start with what Claude can do. If you paste a screenshot of an engineering drawing into Claude, it can attempt to describe what it sees. It might identify that it's looking at a mechanical part, call out some dimensions it can read, and offer general observations about the drawing.
But that's where it stops being useful for real engineering review.
Claude can't open DXF or DWG files. It has no native ability to parse these standard CAD drawing formats. It can't extract the structured data embedded in a drawing - the tolerance values, the GD&T feature control frames, the BOM line items, the revision history in the title block. All of that data exists as structured information in the original file, and Claude can only see pixels in a screenshot.
GD&T interpretation is particularly problematic. A feature control frame contains a geometric characteristic symbol, tolerance value, material condition modifiers, and datum references - all packed into a compact graphical format. Claude might recognize the general shape of a feature control frame from a screenshot, but reliably parsing the specific symbols, modifiers, and datum references? That requires understanding the drawing at a data level, not a pixel level.
BOM tables in drawings contain part numbers, quantities, materials, and descriptions that need to be cross-referenced against the actual assembly structure. Claude sees these as images of tables and might transcribe some text, but it can't verify that the BOM matches the assembly, check for missing components, or validate part numbers against a PDM system.
IN PRACTICE
The technical Q&A feature pulls from real engineering standards with source citations, giving engineers confidence they're getting accurate, relevant answers.
"The technical Q&A feature pulls from real engineering standards with source citations, giving engineers confidence they're getting accurate, relevant answers." - Verified User, Mechanical or Industrial Engineering
Some engineers try to work around Claude's limitations by taking screenshots of their drawings and uploading them. This approach has fundamental issues.
Resolution matters. Engineering drawings contain fine detail - small text, tightly packed GD&T symbols, dimension leaders that overlap, notes in small font sizes. Screenshots compress this information and the AI has to guess at values it can't clearly see. A tolerance of plus or minus 0.005 might look identical to plus or minus 0.008 in a compressed image.
Context is lost entirely. A screenshot of one view doesn't give Claude access to the other views, sections, and details on the same drawing, let alone other sheets in the set. Engineering drawing review requires looking at the full document - checking consistency between views, verifying that section cuts are correctly referenced, confirming that detail views correspond to the correct features.
And perhaps most critically, there's no connection to the 3D model. In modern engineering workflows, the drawing is generated from a 3D CAD model. Reviewing a drawing properly often means checking that the drawing accurately represents the model's geometry, that tolerances make sense given the design intent, and that critical features are properly called out. Claude has no access to any of this context.
Leo AI reads CAD files natively, and that's not a marketing distinction - it's a technical capability built on 3 US patents for reading CAD geometry. Leo can parse B-rep geometry, feature trees, and assembly relationships directly, without requiring screenshots or file conversions.
For engineering drawings specifically, this means Leo can extract and interpret the structured data in the drawing: actual tolerance values, GD&T callouts with correct symbol interpretation, BOM contents, title block metadata, and revision information. It doesn't guess at these values from pixels. It reads them from the data.
Leo can also cross-reference drawing data against the 3D model and against your PDM system. Is the BOM on the drawing consistent with the assembly structure? Does the title block revision match the latest revision in your PDM? Are all critical features properly dimensioned and toleranced? These are the kinds of checks that make drawing review actually useful.
Combined with its training on over one million vetted engineering sources - including ASME Y14.5, ISO standards, and GD&T references - Leo can evaluate whether your GD&T callouts are correctly applied and whether your tolerances align with the manufacturing processes you've specified.
Drawing errors that slip through review have downstream costs that are easy to underestimate. A wrong tolerance on a drawing can result in parts that don't assemble. A missing GD&T callout can lead to a feature being manufactured with inappropriate precision - either too loose, causing functional issues, or too tight, driving up manufacturing cost and scrap rates.
Revision errors - where the drawing doesn't match the latest design revision - are especially expensive. They can result in manufacturing the wrong version of a part, which may not be discovered until assembly or final inspection. At that point, you're looking at scrap, rework, schedule delays, and potentially warranty claims.
Manual drawing reviews catch many of these errors, but they're time-consuming and inconsistent. The thoroughness of a review depends on who's doing it, how much time they have, and how complex the drawing is. AI-assisted review doesn't replace human judgment, but it can flag potential issues that a reviewer should look at, making the process faster and more consistent.
Claude is useful for discussing engineering drawing concepts. If you want to understand GD&T principles, learn about drawing standards, or get explanations of tolerancing approaches, Claude is a solid educational resource.
For actual drawing review - parsing real drawing files, extracting tolerance data, checking GD&T correctness, validating BOMs, and cross-referencing against your 3D models and PDM system - you need AI that can read engineering data natively. That requires purpose-built capability, not a general model looking at screenshots.
FAQ
AI That Reads Your Drawings
Native CAD parsing, not screenshot guessing.
Leo AI reads engineering drawings at the data level - GD&T, tolerances, BOMs, and revision info - with accuracy you can trust for real review work.
Schedule a Demo →
#1 New AI Software Globally - G2 2026
Enterprise-grade security
Trusted by world-class engineering teams
AI That Reads Your Drawings
Native CAD parsing, not screenshot guessing.
Leo AI reads engineering drawings at the data level - GD&T, tolerances, BOMs, and revision info - with accuracy you can trust for real review work.
Schedule a Demo →
#1 New AI Software Globally - G2 2026
Enterprise-grade security
Trusted by world-class engineering teams