Research Engineer - Agent

About the job

Research Engineer – Agent Systems

We (https://www.kaon.io/) build LLM-based Agents for personalized content generation and long-term interactive experiences for millions of users every day. This role focuses on designing, training, and deploying Agent systems with strong memory, personalization, and continual learning capabilities.

Responsibilities

• Design and implement LLM-based Agent architectures for personalized content generation and character interaction, including orchestration workflows, tool use, etc.

• Develop specialized training pipelines using SFT, RFT, DPO, GRPO, and related methods to optimize multi-turn consistency, personalization quality, and memory retrieval accuracy.

• Build realistic Agent training environments (tool-calling sandboxes, user simulators, multi-turn dialogue systems) with well-defined state, action, and feedback spaces.

• Design multi-dimensional reward systems targeting personalization quality, content consistency, memory accuracy, and user satisfaction, leveraging PRM, GRM, and personalized reward modeling.

• Build interpretable and controllable memory architectures that support long-term user modeling, dynamic preference updates, memory versioning, and precise forgetting.

• Implement continual learning mechanisms at the context/token level, evolving Agents through prompt, memory, and tool updates rather than weight changes, while mitigating context rot.

• Design asynchronous “sleep-time” computation mechanisms for memory consolidation, contradiction resolution, abstraction, and retrieval acceleration.

• Run rapid end-to-end experimentation cycles, including evaluation design, data preparation, offline experiments, and online A/B testing, using production metrics to drive continuous iteration.

Requirements

• Strong programming skills in Python and/or C/C++ with solid foundations in data structures and algorithms.

• Hands-on experience with large model training and reinforcement learning.

• Deep understanding of LLM-based Agent systems and practical experience building or training Agents.

• Ability to systematically decompose challenges in long-term memory, personalization, and context management.

• Self-driven and comfortable owning problems from research through production deployment.

Nice to Have

• Publications or research experience in Agent memory systems, personalized generation, continual learning, or context optimization.

• Experience with memory-augmented LLMs, token-space learning, asynchronous compute, or advanced reward design.

• Experience building executable Agent RL environments or user simulation systems.

Compensation: $200,000 – $500,000 total compensation (base + equity), depending on experience and impact.