How we saved an AI giant a million dollars
We built a translator layer for key AI/ML libraries on Apple Silicon
The project
CitadelAI, a leading developer in the artificial intelligence space, ran into a serious problem after migrating to ARM architecture. Their core tooling — including TensorFlow and TensorFlow Data Validation (TFDV) — wasn't originally adapted for that architecture, which created significant friction in their AI-model development and testing workflows.
Solution
IOTA built a reproducible build system that automated the creation of ARM-compatible builds of TensorFlow, TFDV, and TFDV-BSL. This dramatically simplified the deployment and testing of these libraries on next-generation hardware for the CitadelAI team, making the critical tooling readily available and compatible.
Result
Thanks to this initiative, CitadelAI successfully overcame the technical obstacles tied to adopting new hardware and substantially cut development costs. Engineers could now work on building and optimizing AI models without friction, which accelerated the company's innovation cycle. The project showed how an effective solution to a technically complex problem can deliver significant resource savings and lift the technical readiness of the entire organization.
