VCTI Launches Open-Source AI Model for Optical Communications

VCTI has released what it calls the industry’s first open-source AI encoder model built specifically for the optical communications and photonics sectors. The model, VCTI-RoBERTa-Fiber, was unveiled on September 3 and is now freely available on Hugging Face.

Unlike general-purpose AI models trained on internet text, VCTI’s encoder was tuned on more than 2,600 domain-specific documents, including textbooks, IEEE and ITU-T standards, research papers, and technical web content. The company says this specialization makes it better suited for tasks like semantic search, Retrieval-Augmented Generation (RAG) systems, and predictive maintenance applications in fiber networks.

“Most encoder AI models are trained on general internet text—great for everyday language, but not for the precision and depth our industry demands,” said Manuj Nayak, head of VCTI’s AI Center of Excellence, in the announcement. He added that open-sourcing the model is meant to encourage collaboration across the optical ecosystem.

The release is part of VCTI’s broader bet on AI. In April, the company launched its AI Center of Excellence. Earlier this summer, the company also introduced Fiber IQ, a planning tool that integrates Pole IQ and Geology IQ to improve route design accuracy.

The decision to open-source the model follows a growing trend in telecom and networking research, where companies are beginning to share domain-focused AI models to accelerate development; similar moves have been seen in sectors like wireless systems and cybersecurity. VCTI has indicated that a more advanced version trained on a larger body of optical data will follow.

The company says the model offers a ready-to-use tool that could shorten development cycles and cut costs in areas like automated support, intelligent network management, and advanced troubleshooting.

Source: VCTI