F4 Catches Engineering Drawing Errors Before They Become Million-Dollar Manufacturing Mistakes

The Macro: Engineering Drawings Still Get Checked by Hand

Here is something most people outside manufacturing do not realize. Every physical product that gets machined, molded, or assembled starts as an engineering drawing. That drawing contains precise specifications using a language called GD&T, Geometric Dimensioning and Tolerancing, which defines exactly how much a part can deviate from its ideal shape and still function correctly. The ASME Y14.5 standard governs this language and it is dense, technical, and easy to get wrong.

When a drawing has a GD&T error, bad things happen downstream. A machinist reads the drawing, makes the part to spec, and then quality inspection rejects it because the tolerances were impossible to achieve or contradicted each other. The part gets scrapped. The drawing goes back to engineering. The manufacturing line waits. At scale, a single drawing error on a production part can cost hundreds of thousands of dollars in wasted material and lost time.

The current process for catching these errors is manual review. A senior engineer reads the drawing, checks every callout against the standard, verifies that datum references make sense, and flags issues. This takes hours per drawing and depends entirely on the reviewer’s experience and attention span. Some companies use basic checklist tools or drawing comparison software, but nothing that actually understands GD&T semantics and can reason about whether a tolerance scheme is valid.

The market for engineering quality tools is established but stale. Siemens Teamcenter, PTC Windchill, and Dassault’s 3DExperience platform all manage engineering data, but none of them do automated GD&T compliance checking. Anark and TechniGraphics offer drawing-related services but not real-time AI validation. There is a clear gap between the CAD/PLM tools that create drawings and the quality systems that inspect finished parts. Nobody is catching errors at the drawing stage with any kind of automation.

The Micro: Tesla and SpaceX Engineers Fix a Problem They Lived

F4 validates engineering drawings against GD&T standards in real time. You upload a drawing, and the system interprets every GD&T symbol, datum reference, and note, then checks them against ASME Y14.5 (supporting the 1994, 2009, and 2018 variants). It returns a compliance score with specific recommendations for corrections, plus standard definitions, inspection plans, and tolerance analysis.

Paul Shin and Aidan Cantu built F4 and their backgrounds are exactly what you want to see for this kind of product. Paul was a Mechanical Design Engineer at Tesla working on the Drive Unit, where he owned six CyberCab and SEMI production parts. Before that, he led a global heat exchanger program producing over three million units annually. He built liquid-bipropellant rockets at the University of Alabama and UC Berkeley. Aidan was an Avionics Test Engineer at SpaceX, designing hardware for Raptor 3 engine testing. He earned both his B.S. and M.S. in Electrical Engineering in four years, with minors in CS and Math. He co-led the University of Alabama’s Rocket Team to their first liquid rocket engine hotfire. They came through Y Combinator’s Summer 2025 batch.

These are people who have personally dealt with the consequences of bad engineering drawings on production lines that cannot afford delays. The domain expertise is not academic. They have lived inside the pain point.

The competitive position is strong because this is genuinely hard to build. Understanding GD&T well enough to automate compliance checking requires deep knowledge of geometric tolerancing theory, the ability to parse engineering drawings programmatically, and enough manufacturing context to know which errors actually matter in practice. A generic AI company could not build this without years of domain investment. Competitors like Siemens or PTC could theoretically add this to their platforms, but their products are massive enterprise suites that move slowly and charge accordingly.

The Verdict

I think F4 is one of the most defensible products I have seen come out of a recent YC batch. The domain is narrow, the expertise required is deep, and the problem is expensive enough that customers will pay real money for a solution. Manufacturing companies do not haggle over tools that prevent six-figure scrap events.

The risk is market size. GD&T compliance checking is important but it is not a broad horizontal play. The total addressable market is every company that produces engineering drawings, which is large in absolute terms but concentrated among manufacturers, aerospace firms, and defense contractors. Selling to those customers requires enterprise sales motions that take time and patience.

In thirty days, I want to see how accurate the compliance checking is on real production drawings, not demo files. Sixty days, the question is whether engineering teams actually change their workflow to include F4 in the drawing release process or whether it gets used occasionally as a spot-check tool. Ninety days, I want to know whether they have landed a design partner in aerospace or automotive that validates the product at scale. If a company like Lockheed or a Tier 1 automotive supplier starts using F4 as part of their standard drawing review, the product becomes a category. The founding team has the credibility and the technical depth to make this work.