Member of Technical Staff, Architecture & Scaling
Sign up free to see how well your resume matches this role.
What you'll do
- Conduct research on model architecture, optimization, and scaling to improve the capability and efficiency of our largest models.
- Establish strong baselines and run controlled experiments to determine which ideas actually scale to frontier training runs.
- Set training recipes at scale: learning-rate schedules, context length, token budgets, and compute allocation.
- Explore new architectures, including mixture-of-experts, hybrid attention, and long-context extension.
- Diagnose instability in large runs: loss spikes, divergence, numerical issues, and the infrastructure failures that look like research problems.
- Work across modalities, since our models are multimodal from pretraining forward.
What they're looking for
- Hands-on experience training large models, at a scale where compute allocation and stability decisions carry real cost.
- Strong empirical instincts: you design the experiment that distinguishes between two hypotheses instead of the one that confirms the first.
- Fluency with scaling laws and how to use small-scale results to make a frontier-scale bet.
- Deep familiarity with a modern training stack and distributed training across large GPU clusters.
- Strong engineering. Research here means writing the code and reading the profiler, not handing off a spec.
- A record of work that shipped into real models, whether that shows up as papers, systems, or production runs.
Nice to have
- Experience with mixture-of-experts routing, sparse architectures, or long-context methods.
- Work on data mixtures, curriculum, or tokenizer design.
- Kernel-level optimization or mixed-precision training experience.
- Experience with efficiency work aimed at constrained inference targets, including on-device.
Summarised by NextRaise from the employer’s description, which follows in full below.
Full description from employer
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
You'll work on the architecture and scaling of our largest models: what we train, how we train it, and how to spend the next order of magnitude of compute well. This is empirical research with a direct line to production. The recipes you set are the recipes our frontier runs use.
Responsibilities
- Conduct research on model architecture, optimization, and scaling to improve the capability and efficiency of our largest models.
- Establish strong baselines and run controlled experiments to determine which ideas actually scale to frontier training runs.
- Set training recipes at scale: learning-rate schedules, context length, token budgets, and compute allocation.
- Explore new architectures, including mixture-of-experts, hybrid attention, and long-context extension.
- Diagnose instability in large runs: loss spikes, divergence, numerical issues, and the infrastructure failures that look like research problems.
- Work across modalities, since our models are multimodal from pretraining forward.
Requirements
- Hands-on experience training large models, at a scale where compute allocation and stability decisions carry real cost.
- Strong empirical instincts: you design the experiment that distinguishes between two hypotheses instead of the one that confirms the first.
- Fluency with scaling laws and how to use small-scale results to make a frontier-scale bet.
- Deep familiarity with a modern training stack and distributed training across large GPU clusters.
- Strong engineering. Research here means writing the code and reading the profiler, not handing off a spec.
- A record of work that shipped into real models, whether that shows up as papers, systems, or production runs.
Bonus Qualifications
- Experience with mixture-of-experts routing, sparse architectures, or long-context methods.
- Work on data mixtures, curriculum, or tokenizer design.
- Kernel-level optimization or mixed-precision training experience.
- Experience with efficiency work aimed at constrained inference targets, including on-device.
Compensation
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/benefits depending on the specific role. This information will be shared if an employment offer is extended.
Company
Company facts come from this company's own listings. We only show what the postings themselves carry.