Research Scientist Graduate - Conversational AI, PhD
$153.9k - $301k/yr+ Discretionary bonuses/incentives + Restricted stock units
Entry
PhD
Seattle, WA, USA
In Person
AI & Machine LearningApplied Machine Learning, Natural Language Processing & LLMs
About the job
Requirements
- Candidates must be completing or have recently completed a PhD degree in Computer Science, AI, Mathematics, a quantitative discipline, or a related discipline.
- Strong Python skills and proficiency in at least one of C++, Go, Rust, or Java are required.
- Solid machine learning, deep learning, and natural language processing fundamentals are required, along with hands-on experience with large language models or agents through coursework, research, an internship, competition, open source, or a serious side project.
- Basic statistical literacy is required, including computing a confidence interval, explaining what a p-value does and does not mean, and distinguishing a numerical increase from evidence that a system improved.
- The candidate must be able to read a research paper or engineering blog and turn it into working code.
Responsibilities
- Build and operate an agent runtime, orchestrating skills, tools, and context; implementing loop control and intervention, progressive disclosure, and behavior-level guardrails; and building production safety features including pre-flight budgets, timeout truncation, serve-time gates, shadow or swap-in answer delivery, and safe fallback paths.
- Build context and memory capabilities for long multi-turn agents, including structured agentic note-taking, context compaction and summarization, context editing and observation masking, and just-in-time retrieval; manage context as an evolving, itemized playbook with structured diffs and a deterministic curator.
- Develop post-training and data-flywheel systems using supervised fine-tuning, direct preference optimization, and reinforcement learning to internalize rules into model weights, distill to smaller serving models, and turn served conversations into training, evaluation, and retrieval signals.
- Develop tools, skills, and Model Context Protocol capabilities, including tools-as-APIs, connectors, search across large tool inventories, and skill-library governance through description-conflict handling, trigger evaluations, cross-skill misfire matrices, and on-demand loading.
- Build evaluation systems suitable for launch decisions, including large-language-model-as-judge evaluation calibrated to human agreement; paired comparisons, confidence intervals, repeated sampling, pass-at-k; held-out and time-rolling splits with overfitting alarms; and cascaded scoring and cross-family judge panels.
- Build the self-evolving loop from case mining through automatic root-cause analysis, constrained candidate generation, replay verification against frozen regression sets, canary releases, and the flywheel; maintain auditability through a candidate registry with exact runtime read-back, change lineage, and an archive of rejected candidates for future sampling.
- Run online experimentation and causal readouts using shadow, canary, and A/B testing, non-inferiority gates, traffic-split health checks, robust metric definitions, and off-policy counterfactual evaluation when live A/B testing is unavailable.
- Maintain safety and anti-gaming controls by keeping the evaluator and release gate outside the system-editing loop, maintaining never-optimized anchor sets, monitoring full execution traces rather than only final answers, and pairing quality objectives with cost-side constraints.
- Own a high-leverage end-to-end surface and ship it to production across more than 30 languages, measuring performance with business metrics such as customer satisfaction and resolution or containment rate.
Desired Qualifications
- Experience building an agent runtime rather than only calling an API, such as an agent loop or harness, memory or context-management system, retrieval-augmented generation or tool-use agent, or fine-tuned or post-trained model, including at research, hobby, or competition scale.
- Experience with post-training methods including supervised fine-tuning, direct preference optimization, reinforcement learning from human feedback, reinforcement learning from AI feedback, reinforcement learning with verifiable rewards, reward modeling, and reward-hacking defenses.
- Experience with agent systems, including harnesses, context engineering, Model Context Protocol, skills, sub-agents, and tool search.
- Experience with evaluation and experimentation, including large-language-model-as-judge evaluation and calibration, pass-at-k, regression suites, A/B testing, non-inferiority testing, and off-policy evaluation.
- Experience with self-improving or evolutionary systems, including evolutionary program search, candidate archives and parent sampling, automatic prompt or context optimization, multi-objective Pareto selection, and credit assignment.
- Experience with inference and serving technologies such as vLLM, TensorRT-LLM, mixture-of-experts models, key-value or prompt caching, and associated cost engineering.
- PhD publications, strong competition results such as ACM-ICPC, Kaggle, or machine-learning competitions, or notable open-source contributions.
About the company
TikTok is a short-form mobile video platform that allows users to create, discover, and share vertical videos. The app uses an algorithmic feed to surface content personalized to each user, while advertisers can run in-feed ads, branded hashtags, and sponsored challenges, with a Business Center to plan and measure campaigns. The platform differs from competitors through a large global creator community, integrated marketing tools, and rapid trend cycles that drive high engagement. Its goal is to inspire creativity and bring joy by helping people express themselves and giving brands a direct way to reach a broad audience.
Company Size
10,001+
Company Stage
Grant
Total Funding
$740K
Headquarters
Santa Monica, California
Founded
2016
Get referred to TikTok
See people who can refer or advise you
Simplify's Take
What believers are saying
- TikTok Shop generated about $15.8 billion in U.S. sales in 2025.
- October 2026 AI shopping tools and one-click checkout deepen in-app monetization.
- TikTok One Pay's sub-30-day creator payouts strengthen brand and creator supply.
What critics are saying
- August 2026 layoffs cut 325 U.S. roles, weakening moderation and commerce execution.
- The U.S. joint venture still faces national-security scrutiny over algorithm licensing and control.
- A future U.S. divestiture deadline or ban would break TikTok's American growth engine.
What makes TikTok unique
- ByteDance's recommendation engine still drives unmatched discovery and creator reach globally.
- TikTok Shop, One Pay, and Ad Network fuse content, commerce, and advertising.
- Local partnerships with IMDA, Roar AdX, and Converge embed TikTok into markets.
Help us improve and share your feedback! Did you find this helpful?
Benefits
- Health Insurance
- Dental Insurance
- Vision Insurance
- Life Insurance
- 401(k) Retirement Plan
- 401(k) Company Match
- Paid Holidays
- Paid Sick Leave
- Paid Vacation
- Parental Leave
- Wellness Program