Conversational AI Large Model Researcher - Global E-commerce

Role Overview

This position is part of a team focused on creating a leading large-model dialogue system for a massive e-commerce environment. The work supports hundreds of millions of daily active users across the Douyin e-commerce ecosystem, with use cases spanning platform customer service, merchant support, influencer support, and intelligent shopping guidance. The team has already built a comprehensive intelligent conversation solution and continues to improve operational efficiency and customer experience through advanced technology.

Research Goal

The main objective is to build an LLM-powered customer service chatbot for TikTok and Douyin e-commerce. The system should guide conversations end to end, from understanding the customer’s issue to negotiating a resolution and carrying out the final action.

Why This Work Matters

Large language models are well suited to customer service because of their conversational depth and reasoning ability. With the right design, the chatbot can approach the quality of strong human service representatives.

Core Research Focus

The role involves designing a multi-agent architecture around an LLM, with specialized agents handling planning, responses, and tool use. These agents should work together to manage the full service journey, including diagnosing the issue, discussing possible solutions, executing the chosen action, and following up on results.

• The reply agent should produce responses that stay within platform policies and service rules, avoid hallucinations or overly creative answers, and keep the conversation smooth while reaching agreement with the user.
• The planning agent should infer the user’s intent and scenario, retrieve relevant service rules and constraints, and detect any risk-related situations.
• The tool agent should confirm that tool calls are allowed, correctly interpret tool outputs, and manage dependencies between actions.

Key Challenges

One challenge is ensuring every recommendation follows service guidelines, such as refund eligibility windows after parcel arrival and coupon usage limits per customer per week.

Another challenge is balancing strict rule-following with flexibility. A rigid system may provide safe but generic answers, so the reply agent should be able to react to live signals such as user behavior, stated needs, and feedback on proposed solutions. This allows the chatbot to behave more like an effective human service specialist and offer more personalized assistance.

The work also requires self-reflection capabilities, where the model can inspect its own outputs, assess quality, and improve decisions on complex or ambiguous tasks through internal evaluation.

In addition, the system must handle complex image understanding scenarios, including shipping receipts, bank transaction screenshots, product damage photos, and seller qualification documents, all of which may contain important evidence for resolving user issues.

About TikTok

TikTok is a leading short-form video platform with a mission to inspire creativity and bring joy. Its global headquarters are located in Los Angeles and Singapore, and it also has offices in New York City, London, Dublin, Paris, Berlin, Dubai, Jakarta, Seoul, and Tokyo.

Why Join the Company

The organization values curiosity, humility, ownership, and a strong drive to create impact. It encourages experimentation, continuous learning, and a resilient mindset. Teams operate with an “Always Day 1” approach, aiming for meaningful breakthroughs while building together.

Diversity and Inclusion

The company is committed to fostering an inclusive workplace where people are valued for their skills, experience, and perspective. It aims to reflect the diversity of the communities it serves and welcomes a wide range of voices and backgrounds.

Qualifications

• A doctoral degree is required, preferably in computer science, software engineering, artificial intelligence, mathematics, or a closely related discipline.
• A strong grounding in machine learning is expected, including deep learning, multimodal models, and generative modeling, along with strong mathematical ability and a willingness to learn independently.
• Hands-on experience with machine learning and engineering frameworks such as PyTorch or TensorFlow, plus solid programming ability, is necessary.
• Previous work on multimodal large models is preferred, especially in long-text processing or film/television drama-related domains.
• Applicants with publications in leading computer science conferences or journals, or with strong competition experience, will be especially valued.

Additional Information

This role is based in Singapore and is a full-time, on-site position.

The source description does not specify salary, benefits, number of openings, start date, or application deadline.