LLM & Agent Algorithm Graduate (Search) - 2027 Start (PhD)

Overview

The TikTok TT-Search Algorithm & Applied AI team is dedicated to advancing the search experience on TikTok by shifting from traditional "retrieval-and-ranking" methods toward intelligent Agents that proactively complete tasks. This involves developing foundational large language models (LLMs) tailored for search, an Agent execution framework (Harness), multi-agent collaboration mechanisms, and self-improving systems.

Key Responsibilities

• Research and enhance search-specific LLM foundations, including integrating domain knowledge and optimizing post-training pipelines for ultra-long texts, colloquial expressions, and multimodal representations.
• Develop and refine the Agent execution framework, coordinating tool use, planning, memory management, and multi-agent orchestration, including task allocation and communication.
• Design long-term memory systems for LLMs, explore recursive self-improvement techniques, automate training pipelines, and oversee data synthesis and quality assurance.
• Construct automated evaluation and reward frameworks for complex, multi-agent tasks without necessitating manual annotation.
• Apply advanced AI capabilities to business scenarios like long-horizon task Agents, multimodal AI-generated content creation, visual search, on-device intelligence, and ecosystem development.

Qualifications

• PhD in computer science or related fields, either completed or near completion.
• Strong background in machine learning and deep learning, with familiarity of modern LLM and Agent technologies.
• Experience with LLM projects (pre-training, fine-tuning, alignment), Agent development, retrieval-augmented generation (RAG), reinforcement learning, data synthesis, or automated evaluations is advantageous.
• Proficiency in PyTorch or TensorFlow for model training and deployment, understanding distributed and mixed-precision training, model compression, and inference acceleration techniques.
• Knowledge of big data processing frameworks like MapReduce and Spark is a plus.

Preferred Expertise

• Specializations in LLM & Agent technologies: including long-context models, multi-agent orchestration, self-improvement loops, and automated evaluation.
• Competence in computer vision and multimodal learning: image/video retrieval, segmentation, object detection, OCR, graph neural networks, and unsupervised learning; experience in recognized competitions or publications is a bonus.
• Natural language processing skills: pre-training, multilingual understanding, generation tasks, and semi-supervised learning; strong LLM project experience or relevant accolades preferred.
• Strong engineering and collaboration skills coupled with enthusiasm for search and Agent-focused development.

About TikTok

TikTok is a globally recognized platform for short-form mobile video content, striving to inspire creativity and bring joy worldwide. Headquartered in Los Angeles and Singapore, it maintains a diverse presence in multiple international cities. The company promotes a culture grounded in curiosity, humility, resilience, and continuous innovation.

Diversity and Inclusion

TikTok is devoted to fostering an inclusive workplace valuing diverse backgrounds, skills, and perspectives. The company cherishes its global community connections and is dedicated to creating an environment that mirrors its worldwide audience and employees.

Additional Information

• Candidates must be able to commence employment by the end of the year and must clearly state availability and graduation date in their application.