Jobs at Thinking Machines Lab

Build and own internet-scale web-crawling and ingestion systems for pretraining data. Responsibilities include designing distributed crawlers, extraction and deduplication pipelines, data-quality filtering, specialized crawlers, and petabyte-scale infrastructure. The role partners with pretraining teams to assess model impact, improves system reliability and efficiency, and helps define technical direction. Candidates need extensive experience with web crawlers or large-scale data acquisition, distributed systems, and the practical and legal aspects of web data collection.

Design, build, and operate secure sandboxing infrastructure for untrusted, model-generated code and tool calls. Responsibilities include improving isolation with containers and microVMs, developing scheduling and resource-management systems, scaling concurrent executions, adding observability and abuse detection, and owning platform reliability and performance. The role partners with researchers and product teams to provide safe, usable sandboxing primitives.

Lead and build Thinking Machines’ product policy function, establishing usage, content, safety, model release, and developer-facing policies. Hire and manage a team, create product and model risk review frameworks, develop decision-making infrastructure, and partner with Research, Product, Trust & Safety, Communications, Public Policy, and Legal. Translate complex AI policy questions into actionable executive recommendations and responsible launch mitigations.

Lead Thinking Machines’ global public policy strategy and build its policy team. Set positions on frontier AI legislation, safety, disclosure, open-weight models, export controls, and national security. Represent the company before legislators, regulators, and industry groups; manage counsel and coalition relationships; and establish internal policy infrastructure supporting research, product, safety, legal, and executive decision-making across U.S., EU, and international jurisdictions.

Lead the Fine-tuning Science team and set the research agenda for frontier model customization and post-training techniques. Conduct hands-on research in LoRA, parameter-efficient fine-tuning, reinforcement learning, and training stability. Hire and mentor researchers, improve large-scale training reliability and efficiency, collaborate across research, infrastructure, and product teams, and represent findings through publications and community contributions.

Advance fine-tuning and post-training research for frontier AI models, including LoRA and parameter-efficient fine-tuning. Improve the quality, efficiency, stability, and reliability of large-scale fine-tuning and reinforcement-learning runs. Translate research into Tinker’s training defaults, APIs, and open-source cookbook recipes, while communicating findings through papers, technical posts, and community contributions.

Build reinforcement learning and post-training systems for Tinker, including algorithms, data pipelines, numerics, kernels, and distributed training infrastructure. Collaborate with researchers, companies, and external partners; debug real-world RL runs, optimize training pipelines, and enable frontier-model customization. The role requires Python and deep learning framework expertise, with preferred experience in RL stability, low-precision training, LLM serving, scaling studies, and open-source frameworks.

Own full-cycle technical recruiting for infrastructure, systems, networking, and site reliability roles. Partner with engineering leaders on role scoping, sourcing strategies, interview design, talent calibration, candidate assessment, and closing. Build proactive outreach campaigns for specialized infrastructure talent, manage candidate experience, track recruiting metrics, and develop scalable recruiting processes and playbooks.

Lead commercial and product legal work for advanced AI research and products. Draft and negotiate enterprise, data licensing, technology transaction, research collaboration, product partnership, safety, and open source agreements. Advise technical and cross-functional teams, support compliance and product counseling, and build legal infrastructure including playbooks, systems, automation, and outside counsel relationships.

Build and operate engineering infrastructure supporting post-training AI research, including reinforcement learning training systems, sandboxed environments, data pipelines, and agent scaffolding. Embed with research teams, lead projects end to end, improve large-scale systems, debug intermittent failures, and communicate complex technical concepts clearly. The role requires strong Python and engineering fundamentals, autonomy, and experience working in fast-moving codebases.

Develop reliable evaluations and research signals for frontier AI models. Responsibilities include creating capability and usability evaluations, improving grader reliability, auditing benchmarks, building agentic evaluation environments and user simulators, and assessing nuanced behaviors such as preferences, personalization, biases, and values. The role collaborates closely with post-training researchers and engineers and may involve Python, deep learning frameworks, distributed training, and novel evaluation methodology research.

Ensure reliability of large-scale GPU supercomputing clusters by diagnosing hardware, firmware, driver, kernel, and operating-system issues. Investigate root causes, automate fleet monitoring, analyze reliability data, manage firmware qualification and rollouts, coordinate directly with hardware vendors, oversee RMAs, and improve GPU health monitoring. The role also requires writing postmortems and vendor cases and owning reliability initiatives across teams.

Research and develop agentic coding capabilities for frontier AI models. Design reinforcement learning training jobs, coding sandboxes, reward signals, synthetic data pipelines, evaluations, and scalable infrastructure. Analyze large-scale training runs, identify failure modes, and collaborate with infrastructure and post-training teams to ship improvements into model releases.

Conduct AI safety research across data curation, post-training, evaluations, synthetic data generation, and red-teaming. Study how models handle harmful and dual-use requests, develop safety interventions, evaluate long-horizon and agentic behavior, identify jailbreaks and failure modes, and design mitigations. The role requires hands-on Python development, deep learning framework experience, distributed training debugging, and clear technical communication.

Researcher developing and scaling reinforcement learning systems for frontier AI models. Responsibilities include advancing asynchronous RL algorithms, co-designing training recipes and infrastructure, optimizing rollout generation and distributed training, managing large-scale accelerator runs, improving computational efficiency, and conducting rigorous ablations and scaling studies. The role requires full-stack ownership from algorithms through systems implementation and run stability.

Conduct post-training research for frontier AI models by developing and tuning training recipes, designing meaningful evaluations, debugging distributed training, scaling methodologies, and exploring novel datasets. Analyze experimental results, distinguish signal from noise, and publish research, code, datasets, and insights. The role combines theoretical research with high-performance engineering and focuses on alignment, human-AI collaboration, and model usefulness and safety.

Design and optimize distributed infrastructure for large-scale language-model training, focusing on numerical stability, low-precision formats, kernel performance, communication primitives, and parallelism. Build and benchmark systems across multi-GPU and multi-node environments, collaborate on model architectures and training recipes, improve orchestration and monitoring, and share findings through documentation, open-source work, or technical reports.

Supports 3–4 technical leaders by managing calendars, meetings, travel, communications, recruiting coordination, and project commitments. Serves as a key liaison across the company, handles personal logistics, and works autonomously in a fast-changing startup environment. The role requires adapting to different leadership styles, coordinating across time zones, maintaining discretion with sensitive information, and potentially managing a satellite office.

Own the deployment strategy and roadmap for moving trained and fine-tuned AI models into reliable production use. Define serving workflows, autoscaling, monitoring, rollback, incident response, SLAs, pricing inputs, and API/SDK experiences. Partner with infrastructure, research, engineering, GTM, and users to prioritize improvements, resolve dependencies, guide launches, analyze performance and cost tradeoffs, and build a scalable deployment platform.

The Post-Training Product Manager partners with frontier AI researchers to set priorities, connect research with product goals, identify dependencies, and guide decisions. Responsibilities include translating ambiguous questions into learning plans, coordinating research and infrastructure stakeholders, incorporating user and model-behavior evidence, supporting release readiness, documenting decisions, and helping turn research outcomes into usable capabilities.