We wrote translator-layer for key AI/ML libraries for Apple Silicon
CitadelAI, a leading artificial intelligence developer, faced a major challenge after switching to the ARM architecture. Their main tools, including TensorFlow and TensorFlow Data Validation (TFDV), were not originally adapted for this architecture, which created significant difficulties in developing and testing AI models.
IOTA developed a reproducible build system that automated the process of creating ARM-compatible TensorFlow, TFDV, and TFDV-BSL builds. This solution made it much easier for the CitadelAI team to deploy and test the necessary libraries on next-generation equipment, ensuring the availability and compatibility of critical tools.
Thanks to this initiative, CitadelAI has successfully overcome technical obstacles associated with the introduction of new hardware and has significantly reduced development costs. Developers were able to work seamlessly on creating and optimizing AI models, which led to the acceleration of innovation processes in the company. The project demonstrated how effectively solving technically complex problems can significantly save resources and improve technological training across the organization.