Hark

Member of Technical Staff, Mid-training

San Jose · Staff+ · Full-time

Sponsorship not specified$180k-$450kDetected 83 days ago

PythonAlgorithmsMachine LearningPyTorchLLMsAgentic AIA/B TestingRoboticsResearchCommunication

About the role

This role sits at the core of model capability development-defining how data, algorithms, and systems interact to unlock the next frontier of agent behavior.

Design and implement mid-training strategies to improve agent capabilities such as reasoning, planning, tool use, and long-horizon decision-making.
Scale synthetic data generation pipelines (e.g., coding, agent trajectories, multimodal data) and optimize data mixtures to improve downstream RL performance.
Build and optimize distributed training pipelines for large models, ensuring efficiency, stability, and scalability across GPU clusters.
Develop and iterate on evaluation frameworks to measure model capability (e.g., task success, reasoning quality, tool use accuracy) and guide training improvements.
Collaborate cross-functionally with pre-training, post-training, and product teams to align model development with real-world agent use cases.
Hark is an artificial intelligence company building advanced, personalized intelligence.
We're pairing that intelligence with next-generation hardware to create a universal interface between humans and machines.
Experience building or working within simulation or execution environments (e.g., code interpreters, sandboxed execution, game environments, robotics simulators).
Proven ability to design and execute rigorous experiments, with strong intuition for diagnosing training failures and scaling bottlenecks.

Proficiency in Python and PyTorch
comfort working across research and systems code.
Ability to work in a fast-moving, research-forward environment where the right approach is often unknown at the outset.
Proficiency in Python and PyTorch; comfort working across research and systems code.

Experience with mid-training, post-training, or agent-focused model development or coding LLM training.
Familiarity with synthetic data generation, trajectory-based training, or coding/model distillation pipelines.
Experience training or scaling large models (100B+ parameters or equivalent systems).
Contributions to open-source ML systems or publications at top conferences (ICML, NeurIPS, ICLR, ACL, etc.).
Experience optimizing distributed training systems (GPU utilization, memory efficiency, communication).
The pay offered for this position may vary based on several individual factors, including job-related knowledge, skills, and experience.
This information will be shared if an employment offer is extended.

The US base salary range for this full-time position is between $180,000 - $450,000 annually.
The pay offered for this position may vary based on several individual factors, including job-related knowledge, skills, and experience.
The total compensation package may also include additional components and benefits depending on the specific role.
This information will be shared if an employment offer is extended.

One that is proactive, multimodal, and capable of interacting with the world through speech, text, vision, and persistent memory.
Background in reinforcement learning, decision-making systems, or agent frameworks.
Drive technical innovation in areas such as long-context learning, data distillation, and training efficiency, while contributing to the overall model roadmap.

We are looking for a Member of Technical Staff - Mid-Training to lead the development of training strategies that bridge pre-training and post-training, shaping how models acquire reasoning, planning, and tool-use capabilities at scale.

This listing is sourced directly from Hark's careers page and normalized into a canonical job model.