About the Team We build the next-generation unified Agent system for TikTok's global e-commerce customer service — running in 30+ languages across one of the largest e-commerce surfaces on the internet. Our north star is a self-evolving Agent, where post-training, harness, memory / context engineering, tools, and evaluation form one closed loop, and every served conversation becomes the next iteration's training / eval / retrieval / skill-induction signal. We build the agent runtime itself — Codex / Claude-Code-class — not prompts on top of a vendor API. As a new grad, you'll own a real end-to-end piece from day one and ship it to production.
We are looking for talented individuals to join our team. As a graduate, you will get opportunities to pursue bold ideas, tackle complex challenges, and unlock limitless growth.
Successful candidates must be able to commit to an onboarding date by the end of the year. Please state your availability and graduation date clearly in your resume.
Candidates can apply to a maximum of two positions and will be considered for jobs in the order you apply. The application limit is applicable to our Company and its affiliates' jobs globally. Applications will be reviewed on a rolling basis - we encourage you to apply early.
Job Responsibilities
- Build and evolve the agent runtime (harness / agent loop) powering our customer-service agents — orchestrate skills, tools, and context; implement loop control & intervention, progressive disclosure, and behavior-level guardrails.
- Engineer context & memory for long multi-turn agents — agentic memory (structured note-taking), context compaction / summarization, context editing / observation masking, and just-in-time (retrieve-then-load) retrieval.
- Post-train and fine-tune LLMs (SFT / DPO / RL) and build the data flywheel that turns served conversations into training / eval / retrieval signals.
- Design and integrate tools, Skills, and MCP connectors (tools-as-APIs), plus skill / tool search for large tool inventories.
- Build evaluation — LLM-as-judge with human-agreement calibration; regression / safety / cost / latency-aware harnesses; close the offline↔online gap.
- Build the self-evolving loop — replay + user-simulator + Auto-RCA / Auto-GSB / Auto-Policy — so the system continuously improves itself.
- Own one high-leverage end-to-end surface and ship it to production across 30+ languages, measured on real business metrics (CSAT, resolution / containment rate).
Minimum Qualifications
- Individuals who are completing or have recently completed a Bachelor's degree or Master's degree in CS / AI / Math or a quantitative field.
- Strong Python plus one of C++ / Go / Rust / Java
- Solid ML / DL / NLP fundamentals, with genuine hands-on experience with LLMs or agents (coursework, research, internship, competition, open-source, or a serious side project)
- Able to read a paper or an engineering blog and turn it into working code
Preferred Qualifications
- Have built the runtime, not just called an API — even at research / hobby / competition scale: your own agent loop / harness, a memory / context-management system, a RAG or tool-use agent, or a fine-tuned / post-trained model
- Hands-on experience in any one of: post-training (SFT / DPO / RLHF / RLAIF / RLVR, reward modeling); agent systems (harness, context engineering, MCP / Skills, sub-agents, tool search); evaluation (LLM-as-judge, τ-bench / SWE-bench / GAIA / BFCL); inference & serving (vLLM / TensorRT-LLM, MoE, KV / prompt caching); or multilingual NLP — depth in one is enough, breadth welcome