Member of Technical Staff, LLM & Production

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

The Omni team at Hark is building the next generation of AI experiences beyond text, enabling models to understand and generate content across multiple modalities, including text, audio, and vision. Our goal is to create seamless, real-time multimodal intelligence that powers intuitive and immersive user experiences.

We're looking for an engineer who sits at the intersection of model research and production systems. Together with our researchers, you'll decide what "better" means for a model and how to prove it — then build the infrastructure that makes those proofs reliable and repeatable: evaluation harnesses, data pipelines, and model training that hold up at scale and that the whole team can build on.

This is a hands-on, individual-contributor position. The systems that judge whether a model improved, and the environments that drive that improvement, will be yours.

Responsibilities

• Build and own end-to-end evaluation suites for LLM capabilities: task specs, grading logic, and the dashboards researchers check every day.
• Run the eval harness as a production service — sandboxed execution, tool calling, multi-turn and agentic tasks, deterministic replay, and caching — so a full evaluation pass is cheap and gives the same answer twice.
• Own the data pipeline from source to training: collection, filtering, dedup, labeling workflows, synthetic data generation, and the quality gates that catch contamination and distribution drift before anything reaches a run.
• Build RL training environments — reward functions, simulators, verifiers, task generators — and keep them stable as they fan out across many workers.
• Work side by side with researchers: turn a hypothesis into a runnable eval or environment within days, then help make sense of what comes back.
• Pull capabilities out of the product layer and into the model. Find where product teams lean on long prompt instructions or hand-written skills, build the evals and environments that teach the model to do those things natively, and retire the workarounds once it can.