Tencent
Posted 2d ago

优图实验室-大模型Code Agent模型效果评估与优化研究

Tencent
Beijing or Shanghai
OnsiteInternship
Responsibilities
  • building benchmarks
  • optimizing models
  • deploying agents
Requirements
  • Master's or doctoral student in computer science, AI, or software engineering with deep large-model training or Agent framework experience
  • Top-tier research publications preferred
Technical tools mentioned
LLMSFTRLHFRLAIFDPOGRPONLPAPICoding AgentAgent

Job description

课题描述

课题背景:
随着大语言模型(LLM)能力的演进,Code Agent 正超越传统的“编写与调试代码”,升级为能够理解自然语言指令、自动完成综合性任务的“协同助手(Co-work Agent)”。在日常工作与研发场景中,Agent 既需要处理端到端的前端页面开发(如根据需求直接构建管理平台页面、交互组件),也需要应对日常办公协同(如自动化文档生成、复杂表格数据处理、跨系统工作流对接)。腾讯优图实验室希望通过本课题,探索面向复杂代码生成与泛办公协同场景的 Agent 核心技术,打造高效的人机协作体验。
课题挑战:
1. 复杂需求拆解与端到端生成:在前端页面开发或应用构建中,如何让 Agent 精准理解模糊需求,完成从 UI 设计、代码生成到交互逻辑组装的全流程构建。
2. 多模态与异构数据/工具处理:在办公协同(Co-work)场景下,如何让 Agent 灵活调用各类文档、表格及 API 工具,进行精准的数据解析、逻辑推理与自动化操作。
3. 长链路规划与自我修正(Self-Correction):在多步骤复杂任务中,如何提升 Agent 的长流程记忆与规划能力,并根据运行反馈(如报错、排版效果、用户反馈)自动调整策略。
4. 综合能力的客观评估:缺乏能够同时兼顾“代码生成质量”与“日常协同/办公任务完成度”的高精度自动化评测 Benchmark。
具体工作:
1. Code & Co-work 评测体系建设:构建覆盖代码生成(如前端页面/组件构建)与办公协同(如文档/表格处理、任务流自动化)的高质量数据集与评测框架,建立客观的能力诊断机制。
2. 模型训练与优化策略:研究面向多步骤推理与工具调用的 SFT、RLHF/RLAIF(如 DPO/GRPO)及轨迹对齐算法。
3. 前沿技术突破与应用落地:跟踪业界前沿 Agent 技术,推动高可用 Code/Co-work Agent 在腾讯实际产品与业务场景中落地。

课题要求

1、计算机科学、人工智能、软件工程等相关专业的博士及硕士生;
2、有深度大模型训练、Agent框架设计或使用的经历;
3、求知欲驱动的行动力: 对大模型、具身智能及 AGI 的演进具有极其浓厚的兴趣与探索欲。不满足于现有的 SOTA,能够主动追踪并复现全球最前沿的研究进展,对“模型能力的边界在哪里”有持续的好奇心;
4、AI-Native 的工作范式: 擅长利用 Coding Agent 优化工作流,深度 Agent 使用者,能够通过自动化工具极速提升代码质量与实验效率;
5、极致的细节控制力 (Attention to Detail): 在处理大规模数据清洗、多模态对齐及超大规模 RL 训练时,表现出近乎偏执的细心。深信“Garbage in, Garbage out”,能够敏锐察觉实验数据中的细微异常并追溯底层逻辑;
6、坚定的第一性原理思考者: 面对复杂、多模态的混沌问题,不盲从于现有的 Paper 或技术路径。能够拆解问题的本质,从物理底层、信息论或计算效率等基本点出发,推演并设计最优的技术方案;
7、在 NLP/AI/SE 顶级会议或期刊(如 ACL, EMNLP, NeurIPS, ICML, ICLR, AAAI, CVPR, ICSE, FSE, ASE 等)发表过高水平论文者优先。

CSIG — 优图实验室

腾讯优图实验室——离产业最近的AI。我们专注于人工智能领域的前沿技术研究和产业实践落地,涵盖了视觉感知理解、多模态理解、大语言模型、计算加速、AIGC等技术研究领域;作为腾讯顶级的人工智能实验室,优图实验室在多个领域开展技术研究和应用落地,同时为腾讯内外部超500+产品和业务提供技术支持和服务;优图实验室汇聚来自国内外的顶尖人才,并与世界顶级院校和机构合作,我们拥有超过1800项全球专利,1000余篇论文被国际顶会收录。

About Tencent

Multinational technology conglomerate providing internet and entertainment services.

Similar jobs

AI Researcher roles near Beijing, Beijing
5d
Save
Mark Applied
Hide
基于行为奖励与快慢脑协同的实时AI导购研究-阿里星/T-Star
Beijing or Hangzhou
OnsiteInternship
Alibaba
AlibabaNYSE: BABA: Provides online marketplaces, cloud computing, and digital payment services.
Master's or doctorate in computer science, artificial intelligence, machine learning, or related fields; Python/C++, algorithms, RL, deep learning frameworks, distributed training, and AI agent expertise required.
Python, C++, PyTorch, TensorFlow, GitHub, Kaggle
2w
Save
Mark Applied
Hide
微信-多模态AI搜索研究
Beijing or Guangzhou
OnsiteInternship
Tencent
TencentHKEX: 0700: Provides integrated internet services, digital entertainment, and cloud technology.
Master's or PhD in CS or related fields, first‑author papers at top conferences (CVPR/ICCV/ECCV/NeurIPS/ICLR), strong large‑model project or internship experience.
RAG
4mo
Save
Mark Applied
Hide
风控垂域大模型基座和Agentic AI研究-风控
Beijing, Beijing, China
OnsiteFull Time
ByteDance
ByteDance: Developing AI-driven content platforms and mobile applications.
Doctorate (2027) in CS/ML/Math preferred; strong algorithms and data-structure foundation; proficiency in Python/Go/C++; experience in LLM/Agent research and strong independent research ability.
Python, Go, C++
8mo
Save
Mark Applied
Hide
『轻舟▪领航计划』AI人才专项
Beijing or Shanghai or Suzhou
OnsiteFull Time
QCraft
QCraft: Develops autonomous driving solutions for robotaxis and vehicles.
2+ YOEMaster's degree in CS/AI/robotics or related,2+ years AI research or development experience,publications or patents preferred,expertise in large models,perception,or distributed systems.
5d
Save
Mark Applied
Hide
面向AI原生应用的安全智能体关键技术与算法研究-A Star
Beijing or Hangzhou or China
RemoteFull Time
Alibaba Group
Alibaba GroupNYSE: BABA: Global technology specializing in e-commerce and cloud computing.
Research experience in machine learning or security, strong theoretical and practical foundations, programming in at least one listed language, large-model technologies, security knowledge, and strong analytical and collaboration skills.
C++, Java, Python, PEFT, LangChain, LangGraph, LlamaIndex, vLLM, SGLang, CoT, Agent, NeurIPS, ICLR, ICML, CVPR, ICCV, ACL, KDD, S&P, CCS, NDSS, USENIX Security, ACM, CTF
1mo
Save
Mark Applied
Hide
【Ace顶尖实习生】基于agent技术的商业化AI代理研究
Beijing or Shanghai or Hangzhou
OnsiteInternship
Xiaohongshu
Xiaohongshu: Social media and e-commerce platform for lifestyle sharing.
Undergraduate or higher (in-school) in CS/AI/Software Engineering preferred; proficiency with Tensorflow and Pytorch; Linux development experience; skilled in C++, Python, Shell; strong communication and teamwork.
Tensorflow, Pytorch, Linux, C++, Python, Shell
1mo
Save
Mark Applied
Hide
AI 安全研究员
Beijing or Shanghai
OnsiteFull Time
MiniMax
MiniMaxHong Kong Stock Exchange: 0100: Developer of multimodal artificial intelligence foundation models and applications.
Background in computer science, network security, or software engineering; deep practical experience in at least one area (reverse engineering, vulnerability discovery/patching, exploit validation, penetration testing, or security operations) or strong full‑stack engineering and architecture skills; ability to define measurable security problems an...
LLM, Agent, CVE, CTF
2mo
Save
Mark Applied
Hide
顶尖应届-AI天线研究员-手机
Beijing, Beijing, China
OnsiteFull Time
Xiaomi
XiaomiHong Kong Stock Exchange: 1810: Designs and manufactures smartphones, consumer electronics, and electric vehicles.
PhD in antenna or AI-related field preferred; expert in microwave antenna theory and simulation; strong Python and generative/ reinforcement learning skills; publication record and system/agent development experience.
Python, GAN, Diffusion Model, Reinforcement Learning, HFSS, CST