Sr. Machine Learning Engineer I

POS-23565

---

About the Role

HubSpot’s Agent Platform team is building Breeze Studio, our custom agent creation product that lets customers define, deploy, and improve AI agents that work on their behalf inside HubSpot. As a Machine Learning Engineer on the Agent Orchestration ML team, you’ll own the models and systems that determine whether those agents are any good.

This is core ML work: prompt optimization, LLM evaluation, model fine-tuning, and inference infrastructure. You’ll work directly with LLM vendors like OpenAI, run model performance experiments, and ship improvements that customers notice. The team is small (currently 2 MLEs), the surface area is large, and the scope is yours to define.

Custom agents are central to HubSpot’s strategic direction as an AI-first CRM. This is not a supporting role.

What You’ll Do

• Design and run experiments to improve agent quality: better tool use, better reasoning, better outputs, using frameworks like DSPy and VLLM
• Build and maintain evaluation infrastructure to measure what’s working and catch regressions before customers do
• Optimize LLM inference: latency, cost, model routing, and quality tradeoffs
• Partner with product teams on model selection and performance benchmarking
• Work closely with product engineers and PMs to translate customer quality problems into ML hypotheses and solutions
• Own models end-to-end: from research and experimentation to production deployment

What We’re Looking For

• 5+ years of experience in a dedicated ML Engineer role (not ML-adjacent software engineering)
• Strong Python skills; experience with PyTorch, VLLM, DSPy or similar LLM optimization frameworks
• Hands-on experience working with large language models in production: prompt engineering, fine-tuning, evaluation, inference optimization
• Ability to move from ambiguous problem (“this agent isn’t performing well”) to experimental design to shipped improvement
• Comfort working with limited supervision in an early-stage product environment

Why This Role

• Autonomy with impact: Small team, large surface area. Your work directly affects the quality of agents that hundreds of customers build.
• Strategic positioning: Breeze Studio is how HubSpot competes in the AI-native CRM market. Agent quality is the bottleneck while you’ll help own.
• Interesting ML problems: Evaluating open-ended agent behavior, optimizing for latency and quality simultaneously, working at the frontier of LLM capabilities.