About the role#
This role involves building a unified agent system for global e-commerce customer service. You will work on an agent runtime that operates in 30+ languages across a large-scale e-commerce surface. The team focuses on creating a self-evolving agent loop where served conversations provide signals for training, evaluation, retrieval, and skill induction. You will build the agent runtime, evaluation frameworks, and experimentation systems directly, rather than relying on vendor APIs.
What you'll do#
- Orchestrate agent loops, including runtime harness, loop control, intervention, and behavior-level guardrails.
- Build production safety layers such as pre-flight budgets, timeout truncation, serve-time gates, and fallback paths.
- Manage agentic memory through structured note-taking, context compaction, summarization, and retrieval.
- Execute post-training tasks using SFT, DPO, and RL to internalize rules into model weights and distill knowledge into smaller models.
- Govern tools and skills, including API connectors, skill search, and library maintenance.
- Design evaluation systems using LLM-as-judge, human-agreement calibration, and statistical methods like pass^k and confidence intervals.
- Develop self-evolving loops for case mining, root-cause analysis, and replay verification.
- Conduct online experimentation using shadow, canary, and A-B testing, alongside off-policy counterfactual evaluation.
What you'll need#
- Currently pursuing a PhD in Computer Science, Engineering, Operations Research, or a related technical field.
- Proficiency in Python and at least one of C++, Go, Rust, or Java.
- Strong fundamentals in ML, DL, and NLP with hands-on experience in LLMs or agent systems.
- Basic statistical literacy, including the ability to calculate confidence intervals and interpret p-values.
- Ability to translate research papers or engineering blogs into working code.
- Experience in at least one of the following areas: post-training, agent systems, evaluation and experimentation, or self-improving systems.
- A track record of publications, strong competition results in ACM-ICPC or Kaggle, or notable open-source contributions.
- E-commerce or multilingual experience is a plus.
Location & details#
- Location: Seattle, Washington, USA.
- Modality: On-site.
- Employment: Full-time internship.
- Applications are reviewed on a rolling basis.
About TikTok
TikTok operates as an entertainment provider. The company maintains a global presence with offices in cities including Los Angeles, Singapore, New York, London, and Tokyo. It employs over 92,000 people. The organization is privately held.
How to get in at TikTok
Landing an internship at TikTok requires speed because early applicants often get seen before the pile grows. Intern Insider sends an instant alert the moment a role matching your target is published anywhere, so you can apply among the first. This gives you a better chance of your resume being reviewed while the team is still actively filling the position. You should aim to be one of the first people to submit your materials. Reaching out to the right person can also improve your chances of getting a response. Intern Insider surfaces the recruiters behind the company roles so you can contact them directly to ask about the role or a referral. This direct approach often works better than waiting for a response from a general application queue.
