Overview:
SK Telecom, a leading South Korean telecom operator, enhanced its A.(A-Dot) personal AI assistant by deploying multi-agent AI workflows using Microsoft Foundry (formerly Azure AI Foundry). The solution orchestrates specialized AI agents to deliver hyper-personalized, context-aware conversational experiences at telecom scale.
The platform addresses challenges in delivering complex, multi-step tasks (e.g., scheduling, search, media handling) while ensuring compliance with strict data protection and localization laws. It uses fine-tuned models for Korean language nuances and integrates securely with enterprise systems.
Key Features:
- Orchestrates multi-agent workflows that break down and coordinate user goals across specialized agents (scheduling, media, search).
- Manages rich contextual memory with user preferences and retrieval-augmented generation (RAG) for grounded and relevant responses.
- Supports scalable, low-latency performance for millions of users with cost-predictable provisioning (PTU model).
- Integrates voice capabilities via Azure Speech for real-time speech-to-text and text-to-speech across web, mobile, and Teams interfaces.
- Implements responsible AI safety with Azure AI Content Safety and semantic grounding via embeddings and compliance filters.
- Provides fine-tuning of models (e.g., GPT-4.1, GPT-4.1-mini) for domain-specific accuracy and linguistic nuance in Korean.
Results & Impact:
- Scaled to support millions of monthly active users with stable performance, even during traffic peaks.
- Reduced new feature delivery times by three to four months through reusable agentic components and iterative optimization.
- A.(A-Dot) quickly ranked in the Top 10 of the South Korean App Store at launch and grew from ~1.1 M to over 10 M monthly active users within two years.
- Cost savings achieved via Prepaid Token Unit provisioning compared to traditional consumption-based models at telecom scale.
- Enhanced user satisfaction with multi-step, context-aware conversational experiences beyond single-turn interaction models.
AI Technology:
AI Model Types: Generative AI / Large language models (LLMs) such as GPT-4.1 and GPT-4.1-mini, fine-tuned for Korean language and telecom contexts.
AI Purpose: Orchestrate multi-agent collaboration, contextualized reasoning, conversational engagement, voice processing, grounding of responses with RAG.
Application Type: Customer engagement and personalization in telecommunications, conversational AI platforms.
Target Users:
- Telecom customers interacting with the A.(A-Dot) AI assistant.
- SK Telecom AI developers and engineers building and iterating agent workflows.
- Product and feature teams deploying new services across platforms.
- Compliance and security teams ensuring governance and regulatory adherence.
Sources:
- microsoft.com/en/customers/story/25680-sk-telecom-azure-ai-foundry