Member of Technical Staff, Mid-training

About Hark Hark is an artificial intelligence company building advanced, personalized intelligence. One that is proactive, multimodal, and capable of interacting with the world through speech, text, vision, and persistent memory. We're pairing that intelligence with next-generation hardware to create a universal interface between humans and machines.

While today's AI largely operates through chat boxes and decade-old devices, Hark is focused on what comes next: agentic systems that interact naturally with people and the real world. To get there, we're developing multimodal models and next-generation AI hardware together - designed from the ground up as a single, unified interface for a new era of intelligent systems.

About The Role

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

Responsibilities

Design and implement mid-training strategies to improve agent capabilities such as reasoning, planning, tool use, and long-horizon decision-making. , 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.

, task success, reasoning quality, tool use accuracy) and guide training improvements. Conduct rigorous experimentation and ablations to understand training dynamics, scaling behavior, and bottlenecks. Collaborate cross-functionally with pre-training, post-training, and product teams to align model development with real-world agent use cases.

Drive technical innovation in areas such as long-context learning, data distillation, and training efficiency, while contributing to the overall model roadmap.

Requirements

Strong background in machine learning, with hands-on experience training or fine-tuning large models — LLMs, multimodal, or equivalent systems. Deep understanding of reinforcement learning: policy optimization, reward design, exploration, and the interplay between environment design and agent behavior. , 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.

We expect strong candidates to come from a range of backgrounds — RL research, robotics, competitive programming systems, compilers, formal methods, or large-scale ML — rather than post-training specifically. The field is new enough that directly relevant experience is rare; what matters is depth, rigor, and transferability.

Qualifications

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).

Background in reinforcement learning, decision-making systems, or agent frameworks. ). Experience optimizing distributed training systems (GPU utilization, memory efficiency, communication).

Compensation

The US base salary range for this fu