Senior Software Engineer – Global E-Commerce AI Search Infrastructure (TikTok Shop)

About the Team

We're building the next-generation AI search and shopping assistant for TikTok Shop, TikTok's global commerce platform, — powering Q&A cards, in-app chatbot, and visual search experiences that help billions of users discover products, explore stores, and shop through natural conversation. Our team owns the full search stack: from retrieval and ranking to multi-agent LLM engines, post-training infrastructure, and personalized memory. We translate cutting-edge research into production systems at global scale, with a focus on relevance, latency, and fast algorithm iteration.

Responsibilities

• Build AI Search Agents: Design and ship ReAct-based agents with planning, memory, and tool use; implement DAG workflows and RAG pipelines for multi-turn shopping assistance and query understanding. Own the unified Agent Harness across Q&A cards, in-app chatbot, and visual search surfaces — with MCP tool-chain integration and end-to-end A/B support. - Improve LLM Query Understanding: Drive multi-turn conversation, cross-lingual analysis, and LLM reasoning chains for accurate, trustworthy search results; optimize answer generation pipelines (quantization, KV cache, continuous batching) for quality and latency. - Build Personalized Memory Infrastructure: Design high-throughput storage and retrieval for long-term user profiles and real-time memory (MemAgent); build near-real-time feature pipelines over billion-scale behavior sequences for low-latency serving of personalized signals. - Contribute to Training and Inference Infrastructure: Collaborate on post-training pipelines (SFT, RL, distillation) and model-serving infrastructure (tensor parallelism, speculative decoding, PD separation) to accelerate experimentation and hit latency targets. - Ship Research to Production: Bridge research and engineering — partner with algorithm teams to evaluate agent and LLM innovations, accelerate adoption, and ensure new capabilities land stably in production at scale.