Machine Learning Engineer Graduate (Conversational AI)

Machine Learning Engineer Graduate (Conversational AI) at TikTok (Seattle) for a 2027 start, building a self-evolving unified agent system for e-Commerce customer service across 30+ languages.

Responsibilities

• Agent runtime (harness / agent loop): Orchestrate skills, tools, and context; implement loop control and intervention; use progressive disclosure and behavior-level guardrails.
• Production safety layer: Build pre-flight budgets and timeout truncation; add serve-time gates; deliver shadow and swap-in answers; define safe fallback paths.
• Context and memory for long multi-turn agents: Implement agentic memory via structured note-taking; perform context compaction/summarization; support context editing and observation masking; use retrieve-then-load (just-in-time) retrieval.
• Post-training and the data flywheel: Apply SFT, DPO, and RL approaches to internalize rules into weights so the prompt gets shorter; distill to smaller serving models; convert served conversations into training, evaluation, and retrieval signals.
• Tools, skills, and MCP: Develop tools-as-APIs and connectors; enable skill/tool search across large inventories; govern a skill library (description conflict handling, trigger evaluations, cross-skill mis-fire matrices); load skills on demand instead of dumping all definitions into context.
• Evaluation you can bet a launch on: Use LLM-as-judge with human-agreement calibration; apply statistical rigor (paired comparison, confidence intervals, repeated sampling, pass^k); maintain held-out and time-rolling evaluation splits with overfitting alarms; use cascaded scoring and cross-family judge panels to reduce evaluation cost at scale.
• Self-evolving loop: Run case mining → automatic root-cause → constrained candidate generation → replay verification against frozen regression sets → canary → data flywheel.
• Auditable system plumbing: Maintain a candidate registry with exact runtime read-back, change lineage, and an archive of rejected candidates for sampling in later rounds.
• Online experimentation and causal readout: Execute shadow/canary/A-B testing; enforce non-inferiority gates; monitor traffic-split health; define metrics that hold up under scrutiny; apply off-policy counterfactual evaluation when live A/B is not possible.

Requirements

• Completed or recently completed a Bachelor’s degree in Computer Science, Electrical Engineering, Mathematics, Statistics, or a related discipline.
• Strong Python plus one of: C++, Go, Rust, or Java.
• Solid ML/DL/NLP fundamentals with hands-on experience using LLMs or agents through coursework, research, internship, competition, open-source, or a serious side project.
• Basic statistical literacy: compute confidence intervals, explain what a p-value does and does not mean, and distinguish “number went up” from “system got better.”
• Ability to read a paper or engineering blog and implement working code.

Technologies

• Python, C++, Go, Rust, Java
• SFT, DPO, RLHF, RLAIF, RLVR
• vLLM, TensorRT-LLM, MoE, KV
• MCP
• A/B, LLM-as-judge
• ACM-ICPC, Kaggle

Preferred Qualifications

• Built agent runtime functionality beyond calling APIs, even at research/hobby/competition scale (agent loop/harness, memory/context management, RAG or tool-use agent, or fine-tuned/post-trained model).
• Hands-on depth in one of the following areas (breadth welcome, depth in one sufficient): post-training (SFT/DPO/RLHF/RLAIF/RLVR, reward modeling, reward hacking and defenses); agent systems (harness, context engineering, MCP/skills, sub-agents, tool search); evaluation/experimentation (LLM-as-judge and calibration, pass^k, regression suites, A/B and non-inferiority testing, off-policy evaluation); self-improving/evolutionary systems (evolutionary search, candidate archives/parent sampling, automatic prompt/context optimization, Pareto selection, credit assignment); inference/serving (vLLM/TensorRT-LLM, MoE, KV/prompt caching, cost engineering); multilingual NLP.
• Publications (for PhD), strong competition outcomes (ACM-ICPC / Kaggle / ML competitions), or notable open-source contributions.
• Taste for accuracy in metrics and significance, including willingness to communicate when results are not yet significant.
• e-Commerce or multilingual experience is a plus (not required).

Benefits

• Day one access to medical, dental, and vision insurance.
• 401(k) savings plan with company match.
• Paid parental leave.
• Short-term and long-term disability coverage.
• Life insurance.
• Wellbeing benefits.
• 10 paid holidays per year.
• 10 paid sick days per year.
• 17 days of Paid Personal Time (prorated upon hire with increasing accruals by tenure).

Compensation

• Base salary range (Seattle): $121,600 - $243,200 per year.
• Compensation may vary outside the range based on qualifications, skills, competencies, experience, and location.
• Base pay is part of the Total Package and the role may be eligible for additional discretionary bonuses/incentives and restricted stock units.

Additional Application Constraints

• Must be able to commit to an onboarding date by the end of the year.
• State availability and graduation date clearly in the resume.
• Apply to a maximum of two positions; applications are considered in the order submitted.
• Application limit applies globally for the company and its affiliates.
• Reviewed on a rolling basis; apply early.

Location and Role Details

• Location: Seattle, WA (onsite).
• Employment type: Regular.
• Job code: A31120.

Team Introduction

• Build a next-generation unified Agent system for TikTok’s global e-Commerce customer service, running in 30+ languages.
• North star is a self-evolving agent where post-training, harness, memory/context engineering, tools, and evaluation form a closed loop and each served conversation becomes the next iteration’s training/eval/retrieval/skill-induction signal.
• Evaluation and experimentation are treated as first-class systems.
• Develop the agent runtime (not prompts on top of a vendor API).
• Work includes generative recommendation, large recommendation models, multimodal representation learning, and cross-domain value modeling.

Accommodation and Fair Chance Notices

• TikTok provides reasonable accommodations in recruitment for candidates with disabilities, pregnancy, sincerely held religious beliefs, or other reasons protected by applicable laws. Request support at https://tinyurl.com/RA-request.
• For Los Angeles County (Unincorporated) candidates: qualified applicants with arrest or conviction records will be considered in accordance with applicable fair chance laws; the company notes that criminal history may have a direct adverse relationship on duties including interacting with internal/external clients/colleagues (occasionally unsupervised), handling confidential information and access to IT systems, and exercising sound judgment.