Rostelecom has initiated a market research process to procure a large-scale AI platform for analyzing voice calls and services. The system is designed to handle traffic for 3 million subscribers, supporting up to 30,000 concurrent calls and 1,200 new calls per second. The platform must integrate with the state-run Anti-Fraud system to detect scams in real time.
The technical requirements specify that the platform must monitor various communication channels, including 2G/3G networks, VoLTE, VoWiFi, and the Russian messengers Maks and VK. The system is expected to utilize machine learning models for speech recognition, emotion detection, deepfake identification, and speaker verification, with model retraining scheduled every two weeks.
The hardware requirements mandate high-performance ML servers with GPU support and x86_64 Softswitch servers registered with the Ministry of Digital Development. The infrastructure must ensure a call completion success rate of at least 96%. The project covers deployment across several major Russian cities, including Moscow, Saint Petersburg, and Yekaterinburg.
Representatives from VK have stated that the requirements outlined by Rostelecom are technically unfeasible for their products. Industry experts note that while call analysis is common in the telecom sector, extending such monitoring to encrypted messenger traffic represents a significant shift in regulatory and technical approaches to communication security.
