How Tencent Uses RAG, GraphRAG, and Agents to Power Large Language Model Applications

Overview

In this article we explore how Tencent applies its large language models to a wide range of business scenarios, highlighting the use of Retrieval‑Augmented Generation (RAG), GraphRAG for role‑playing, and Agent technology for goal‑driven tasks.

Core Application Scenarios

Content Generation: ad copy, comment assistance, etc.

Content Understanding: text moderation, fraud detection.

Intelligent Customer Service: knowledge Q&A, user guidance.

Development Copilot: automated code review, test‑case generation.

Role‑Playing: intelligent NPC interaction in games.

Key Enabling Technologies

Supervised Fine‑Tuning (SFT)

Fine‑tunes a base large language model with domain‑specific data, embedding business knowledge to enable targeted task responses.

Retrieval‑Augmented Generation (RAG)

Integrates external knowledge bases and retrieval mechanisms into the generation process, improving explainability and significantly reducing hallucinations; commonly used in smart customer service and document assistants.

Agent (Intelligent Agent)

Leverages external tools so the model can perform multi‑step reasoning, planning, and execution, making it suitable for complex tasks that require sequential decision‑making.