MegaFon: assistente IA 'Eva' bloqueia bilhões de chamadas spam e ML otimiza estações rádio-base
The 'Eva' AI assistant protects tens of millions of subscribers and blocks or intercepts billions of spam and scam calls annually—detecting newly rotated fraud numbers within minutes of campaign initiation.
| Company & Sector | МегаФон (MegaFon) — Telecomunicações |
|---|---|
| Models & AI Stack | Signaling Traffic ML · Conversational Voice AI (ASR/NLU/TTS) · Self-Organizing Network (SON) · Cell Load Forecasting · Real-Time Stream Processing · Predictive Maintenance |
| Cronologia | 2021–2022 — rollout of SON radio optimization and launch of 'Eva' voice assistant; 2023–2024 — scaling 'Eva' AI spam protection to tens of millions of subscribers and ML base-station power saving; 2025–2026 — generative LLM workflows across customer care and network engineering. |
| Verification | Verificado 2026-10 · 3 primary sources |
Contexto e escala do negócio
MegaFon is one of Russia's largest nationwide mobile operators, serving over 76 million subscribers across more than 200,000 2G/3G/4G/LTE base stations.
In modern telecommunications, AI operates on two fronts simultaneously: autonomous radio access network management (load balancing, predictive fault detection, and power saving across tens of thousands of cell towers) and subscriber protection against voice fraud and spam.
Between 2023 and 2025, MegaFon's 'Eva' AI voice secretary and anti-spam suite became one of the most widely adopted consumer AI products in the telecom market.
Problema e gargalos operacionais
The first massive challenge is voice spam and social-engineering fraud. Scammers rotate ephemeral virtual numbers and SIM banks rapidly, rendering static number blacklists obsolete within minutes of a campaign launch.
Second is Radio Access Network (RAN) energy consumption and configuration complexity. Cell towers are provisioned for peak daytime traffic, burning megawatts of electricity overnight when extra frequency carriers sit idle.
Third is handling tens of millions of subscriber support inquiries across voice and chat channels without wait times during regional outages.
Arquitetura e implementação de IA
Inside the 'Eva' protection suite, MegaFon combined real-time behavioral ML over network signaling metadata (call duration distributions, cell-tower handoff anomalies, burst-dialing signatures within seconds of campaign start) with conversational speech AI (ASR + NLU + TTS).
High-confidence spam calls are blocked at the switch before the handset rings, while unknown or unanswered calls are answered by 'Eva'—an AI voice agent that converses naturally with callers, identifies couriers vs. telemarketers, and delivers a real-time transcript to the subscriber.
At the infrastructure layer, Self-Organizing Network (SON) ML models forecast traffic per cell sector, dynamically balancing load across neighboring antennas and placing underutilized frequency carriers into deep sleep modes during off-peak hours.
“"A mobile carrier sees the signaling signature of a robo-calling wave at the network switch before the first handset even rings. Combining network-level ML with an AI voice secretary protects subscribers natively without third-party app installs." — MegaFon Technology
Impacto mensurável e economia
The 'Eva' AI assistant protects tens of millions of subscribers and blocks or intercepts billions of spam and scam calls annually—detecting newly rotated fraud numbers within minutes of campaign initiation.
Intelligent SON traffic balancing and ML power-saving modes cut base-station electricity consumption by 7–12% across deployed sectors without degrading mobile data speeds, saving hundreds of millions of rubles in utility costs.
In customer care, conversational AI resolves 60–70%+ of routine tariff, billing, and network-diagnostics inquiries without human agents.
Lições para a indústria
Network-level signaling ML detects new spam numbers within minutes of campaign launch—hours before user-reported blacklists react.
An AI voice secretary that sends live call transcripts eliminates subscriber anxiety about ignoring unknown numbers from couriers or clinics.
Predictive ML carrier sleep modes across 200,000+ base stations yield massive direct utility cost reductions with zero QoS impact.
Embedding AI natively at the core network switch achieves vastly higher adoption than standalone downloadable utility apps.
Fontes primárias verificadas
Perguntas frequentes sobre a notícia
How does the AI distinguish a real delivery courier calling from a new number vs. a spam bot?
Network signaling ML checks calling velocity and cell movement patterns, while the 'Eva' voice agent uses NLU to recognize delivery-related speech and forwards a transcript immediately.
Does mobile coverage degrade at night when ML puts frequency carriers to sleep?
No. Baseline coverage carriers remain active 24/7; only supplemental capacity layers used for daytime peaks sleep, waking automatically within seconds if traffic rises.
Do subscribers need to install a separate app for 'Eva' to work?
No. Call interception and the AI voice secretary execute directly on MegaFon's core network switches, functioning even when the handset is offline.
Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews
Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.
O essencial da IA — uma vez por semana
Sete histórias que realmente importaram, escolhidas a dedo. Sem ruído nem releases.
Pronto! Verifique seu e-mail para a confirmação.