How Tencent’s LLM Powers Real‑World AI: From RAG to Agents

Overview

In this article we explore Tencent’s large language model (LLM) applications across various business scenarios, focusing on how cutting‑edge techniques improve model intelligence and user experience.

Key application scenarios

Content generation : e.g., ad copy, comment assistance.

Content understanding : e.g., 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.

Core technologies

Tencent employs three main techniques to deploy LLMs:

(1) Supervised Fine‑Tuning (SFT)

Fine‑tunes a base model with domain‑specific data, embedding business knowledge directly into the model for targeted task handling.

(2) Retrieval‑Augmented Generation (RAG)

Combines external knowledge bases and retrieval mechanisms with generation, enhancing explainability and reducing hallucinations in use cases such as intelligent customer service and document assistants.

(3) Agent (Intelligent Agent)

Integrates external tools, enabling the model to perform multi‑step reasoning, planning, and execution for complex tasks.

Through these combined approaches, Tencent drives intelligent, efficient solutions across content creation, understanding, development assistance, and interactive role‑playing.