Senior Algorithm Engineer – Customer Service Chatbot AI Agent

About The Team

The Marketplace Intelligence and Data team at Shopee focuses on creating sustainable and efficient data and intelligence tools that boost Shopee's business growth. Their responsibilities include managing the e-commerce data warehouse, developing merchant and operations data products, and creating algorithms for products and marketing. Additionally, they handle foundational AI technologies such as Machine Translation, Speech, Image Algorithm, and Real-person Authentication.

The Customer Service Chatbot team is developing advanced multilingual and multi-scenario customer service systems for both consumers and merchants. Their goal is to implement cutting-edge technologies like large language models (LLMs), AI agents, reinforcement learning (RL), and others to enhance real-world customer service applications. They continuously improve the AI models' decision-making, understanding, planning, and execution through Post-Training, Agentic RL, Harness, and Loop Engineering.

They have launched an AI agent-based customer service system and aim to evolve it into an intelligent agent capable of context comprehension, planning, autonomous execution, and continuous self-enhancement, ultimately outperforming human service agents.

Key Responsibilities

• Engage in or lead development efforts around AI Agent architecture, Harness, Loop Engineering, Post-Training, and Agentic Reinforcement Learning.
• Design and refine AI Agent architectures focusing on planning, tool integration, memory management, and multi-agent coordination.
• Develop toolsets, context engineering methods, SOP and knowledge integration to maximize foundation model use in customer service contexts.
• Boost agent system performance by implementing model layering, parallel scheduling, context compression, and cache reuse to minimize latency and inference costs.
• Create a continuous self-evolving loop for the system that leverages real-time customer feedback and automated evaluation to generate new training data for Post-Training and RL, followed by gray-scale validation.
• Establish an automated evaluation system including LLM-as-a-Judge frameworks to benchmark performance with offline and business metrics.
• Formulate post-training strategies tailored to customer service scenarios such as supervised fine-tuning (SFT), Direct Preference Optimization (DPO), GRPO, and Proximal Policy Optimization (PPO), using both explicit and implicit feedback signals.
• Develop Agentic RL objectives focused on decision points like action selection, tool usage, and dialogue strategies, exploring reinforcement learning from human feedback (RLHF) and AI feedback (RLAIF).
• Stay abreast of latest research and drive integration of advanced Post-Training, RL, Agent, Harness, and Loop Engineering practices into production.

Required Qualifications

• Master’s degree or higher in Computer Science, Artificial Intelligence, or related discipline.
• Minimum 3 years of experience in Algorithm Engineering or Machine Learning.
• Strong foundation in computer science, mathematics, algorithms, and data structures.
• Deep knowledge of machine learning and deep learning fundamentals, with particular expertise in Transformer and large language model architectures.
• Hands-on research or practical experience in one or more of these areas: post-training (SFT, DPO, GRPO), reinforcement learning including agentic RL, direction mechanisms like planning and reasoning, or harness/loop engineering frameworks.
• Proficient programming skills in Python along with familiarity in deep learning frameworks like PyTorch or TensorFlow.
• Interest in interdisciplinary challenges combining research, engineering, and business applications, ready to test and improve approaches in live environments.

Preferred Qualifications

• Previous involvement in AI Agent or large language model projects within renowned AI organizations.
• Experience designing scalable AI Agent systems and improving inference efficiency via model distillation, routing, parallel scheduling, and context engineering.
• Publications in prominent conferences or journals relating to NLP, LLM, RL, or Agent technologies.
• Direct experience working on LLM post-training or agentic reinforcement learning projects.
• Practical application of RLHF or agentic RL in dialogue, customer service, or Agent contexts.
• Knowledge of distributed training techniques and large model optimization frameworks like DeepSpeed or Megatron.