LLM & Agent Algorithm Graduate (Search) - 2027 Start (PhD)
Core
Building the search-domain LLM foundation and Agent execution framework (Harness) to shift search from retrieval-and-ranking to proactive task completion.
Role type
PhD-level research scientist (LLM & Agent algorithms)
Builds
Search LLMs, multi-agent orchestration systems, self-improvement loops, and multimodal AIGC features for TikTok Search.
Domain
Search, Large Language Models, Multi-Agent Systems, Multimodal AI
Deliverable
production ML models
Required skills
Machine learning, deep learning, LLM pre-training/fine-tuning/alignment, multi-agent orchestration, automated evaluation, PyTorch, distributed training, model compression/inference acceleration, big data frameworks
Preferred skills
Publications at NeurIPS/ICML/ICLR/ACL/EMNLP/CVPR/ICCV/ECCV/AAAI, experience with RAG/RLHF/RLAIF, computer vision/multimodal learning, natural language understanding/generation
Technologies
PyTorch, TensorFlow, MapReduce, Spark, TensorRT, quantization, pruning, distillation
Responsibilities
Develop post-training pipelines for search LLMs (ultra-long-text, multimodal), build multi-agent collaboration algorithms and scheduling, implement long-term memory and self-improvement mechanisms, construct automated evaluation and reward systems, implement long-horizon task Agents and multimodal AIGC features.
Seniority
PhD, Research Scientist
