# Peter Morales - Code Metal Source: https://www.youtube.com/watch?v=4GgMvA7FGDc Recap page: https://rapidrecap.app/video/4GgMvA7FGDc Generated: 2026-02-04T22:03:43.84+00:00 --- ## Quick Overview Code Metal's approach focuses on creating automated, profile-driven performance optimization workflows that reliably translate and target code across diverse hardware, specifically addressing the limitations of current AI software tools in mission-critical systems by providing rigorous verification before deployment. **Key Points:** - Current AI software tools fail in mission-critical systems due to a lack of validation, open-ended inputs, and stochastic outputs. - Code Metal provides a high-code solution (not low-code/no-code) to automate workflows that help teams reliably automate code verification and profiling before deployment. - Use cases for Code Metal include Rapid Development (fast iteration), Code Portability (hardware retargeting), and Code Modernization (source-to-source translation). - The company has FY25 projections of $15M and is working with 7 Fortune 500 companies across Defense, Industrial, and Semiconductor sectors. - Examples of code pipelines show translation from Python/Matlab to C++/OpenCL/Rust/eBPF/Synthesizable RTL for various uses like HPC, chip design, and safety systems. - The company recently raised a Series A round with Excel and is hiring engineers interested in the intersection of Formal Methods, Compiler Design, and AI. ![Screenshot at 00:00: Speaker presenting a slide titled 'Why LLMs Fail in Production' which lists key failure points: 'No validation', 'Open-ended inputs', and 'Stochastic outputs', setting up the problem Code Metal aims to solve.](https://ss.rapidrecap.app/screens/4GgMvA7FGDc/00-00-00.jpg) **Context:** The speaker presents Code Metal, a platform designed to address the unique challenges of software development in mission-critical industries (like Defense, Automotive, Industrial Automation, and Semiconductor) where software must be correct the first time, unlike typical AI applications. The talk contrasts the inherent risks of LLMs in these environments (no validation, stochastic outputs) with Code Metal's high-code, verifiable approach to code translation and optimization across different hardware targets. ## Detailed Analysis The presentation argues that existing AI software tools are unsuitable for mission-critical systems because they lack validation, accept open-ended inputs, and produce stochastic outputs, contrasting with industries like defense or automotive that demand absolute correctness upon initial deployment (0:00-0:14). The speaker recounts experiences at Microsoft working on the F-35 where code handoffs required deep domain knowledge, highlighting the cost of errors (0:15-0:28). Code Metal's approach is presented as a high-code solution that automates the workflow from code to metal, enabling reliable, systematic code verification and profiling across different hardware targets (2:13-2:37). This is demonstrated through use cases like Rapid Development (Fast Iteration), Code Portability (Hardware Retargeting), and Code Modernization (Source-to-Source Translation) (3:04-3:35). Specific examples show code moving from Python/Matlab/CUDA/etc. to destinations like C++/OpenCL/Rust/eBPF, serving critical functions like HPC, embedded systems, and chip design (3:48-3:53). The company shows traction with FY25 projections of $15M and partnerships with 7 Fortune 500 companies across key sectors (3:54-4:10). Finally, the speaker mentions raising a Series A round with Excel and actively hiring engineers skilled in formal methods, compilers, design, and AI (4:17-4:33). ### Why LLMs Fail in Production - No validation - Open-ended inputs - Stochastic outputs - AI software tools miss real problems in mission-critical systems ### The Problem - Mission Critical Engineering and manufacturing industries demand software that gets it right the first time (e.g., Defense, Automotive, Industrial Automation, Semiconductor companies like Lockheed Martin, RTX, Bosch, AMD). ### Our Approach (Code Metal) - Automates workflows that help teams reliably automate code verification and profiling for development, testing, management, and deployment across various languages (Python, Rust, C++) and targets (C++, OpenCL, Rust, eBPF, RTL). ### Use Cases - Rapid Development (Fast Iteration) - Code Portability (Hardware Retargeting) - Code Modernization (Source to Source Translation). ### Traction - FY25 projection of $15M - 7 Fortune 500 customers deployed across Defense, Industrial, and Semiconductor sectors. ### Hiring - Seeking engineers interested in the intersection of Formal Methods, Compiler Design, and AI; company recently raised Series A with Excel. ![Screenshot at 00:00: Speaker presenting a slide titled 'Why LLMs Fail in Production' which lists key failure points: 'No validation', 'Open-ended inputs', and 'Stochastic outputs'.](https://ss.rapidrecap.app/screens/4GgMvA7FGDc/00-00-00.jpg) ![Screenshot at 01:52: Slide illustrating the problem: 'Mission Critical Engineering and manufacturing industries demand software that gets it right the first time,' with examples of mission-critical industries shown in four quadrants.](https://ss.rapidrecap.app/screens/4GgMvA7FGDc/00-01-52.jpg) ![Screenshot at 02:12: Diagram detailing Code Metal's approach, showing Python/Rust/CUDA inputs feeding into the Code Metal platform, which handles Compilation, Optimization, and Verification before deployment.](https://ss.rapidrecap.app/screens/4GgMvA7FGDc/00-02-12.jpg) ![Screenshot at 03:36: Slide detailing 'Automated Profile-Driven Performance Optimization' showing the lift in productivity achieved by moving from proprietary MATLAB code to a Code Metal solution targeting HDL.](https://ss.rapidrecap.app/screens/4GgMvA7FGDc/00-03-36.jpg) ![Screenshot at 03:48: Table titled 'Example Code Metal Pipelines' showing various language origins \(Python, CUDA, C++\) mapping to destinations \(C++, OpenCL, Rust\) for specific use cases like HPC, safety systems, and chip design.](https://ss.rapidrecap.app/screens/4GgMvA7FGDc/00-03-48.jpg)