Research Scientist, Post-Training and Agentic Systems

About Hextrion

Hextrion is developing next-generation semiconductor advanced packaging to dramatically reduce the energy use, cost and physical footprint of future AI compute. As transistor scaling reaches its physical limits, advanced packaging has become the primary driver of AI performance, determining how efficiently data and power move to and between chips.

We are led by former C-suite and senior executives from the global advanced packaging leaders serving NVIDIA, AMD, Broadcom, Google, Amazon and Microsoft. They are joined by physics AI researchers from Harvard and industry leaders who have deployed production AI systems at TSMC and Apple component manufacturers. We combine deep packaging expertise with physics AI agents and reinforcement learning to design and manufacture advanced packaging with greater performance, precision and yield than current methods deliver.

About the role

As a Research Scientist on Post-Training and Agentic Systems, you'll post-train specialized language models with deep knowledge of advanced packaging design and manufacturing, and you'll build the agents that use those models together with our physics AI models and engineering tools to design packages and optimize how they're manufactured. Our physics AI models and engineering tools provide fast, accurate feedback on how designs and processes will behave in the physical world, serving both as tools for agents and as verifiers for training.

What you'll do

Post-training specialized language models

• Curate and generate high-quality training data from engineering knowledge, design data and manufacturing data, including synthetic data.
• Post-train language models through supervised fine-tuning, reinforcement learning and distillation to build deep expertise in advanced packaging design and manufacturing.
• Use physics AI models, simulators and engineering checks as verifiable rewards for reinforcement learning.
• Build evaluations that measure model performance on real design and manufacturing tasks, and make sure models are efficient and secure to deploy.

Building agentic systems

• Build design agents that explore and optimize advanced package designs using physics AI models, simulators and electronic design automation (EDA) tools.
• Build manufacturing agents that optimize processes, diagnose issues and support engineers on the production line.
• Train agents with reinforcement learning, using digital twins and engineering tools as environments.
• Develop tools and interfaces that let agents reliably use simulators, optimizers and engineering software.
• Enable models and agents to learn continuously from live design and manufacturing data, with safeguards for real-world deployment.

What we're looking for

• A track record of exceptional work, such as first-author publications at top venues, or building and shipping LLM, agent or RL systems at a leading AI lab or tech company.
• Hands-on experience post-training LLMs, including supervised fine-tuning, reinforcement learning (for example with verifiable rewards) and distillation.
• Experience building LLM-based agents, tool-using systems or reinforcement learning environments.
• Experience designing evaluations and training data for complex, real-world tasks.
• Strong engineering skills and the ability to take systems from research to deployment.
• High ownership and the drive to tackle ambiguous, real-world problems end to end.
• A plus: experience training domain-specific models, applying AI to science, engineering, semiconductors, EDA or manufacturing, or working with physics-based simulation.

Role Details

• Minimum education: PhD or equivalent industry experience.
• Compensation: $180,000–$350,000 + equity, depending on experience
• Visa sponsorship: Yes. We will support the visa process.