Description
As a Software Engineer on the Model Lifecycle and Signals team at SpaceXAI, you will own the bridge between frontier models and every user-facing Cursor product, used at a scale few teams ever see.
You will decide when to use which model, spot where models go off track, and build the guardrails to keep them on track, and shape what the next generation of models should learn.
This role sits at the intersection of infrastructure, data, and product.
You will bring models from preview to production, dig into real-world data to find out what's actually working, and partner with research, inference, product, and data science teams to turn that evidence into better products and better models.
Every improvement you make compounds across our products and every model we ship.
Responsibilities
- Get new frontier models into users' hands the day they're ready, and retire old ones without anyone noticing.
- Evaluate internal and external preview models before launch and give meaningful feedback to the teams building them.
- Build Cursor Router: decide which model handles each request, balancing quality, speed, and cost at massive scale.
- Build the eyes and ears of the product, and the experimentation platform that puts them to work: signals that show where each model shines, where it goes off track, and what 'better' actually means.
Requirements
- You are an exceptional engineer with a track record of shipping production systems people rely on.
- You own problems end to end, and you'd rather automate a painful process than repeat it.
- You've built and operated evaluation, measurement, or model serving systems, such as AI evals, experimentation, ranking/relevance, or routers.
- You can turn ambiguous 'quality' questions into concrete metrics, pipelines, and decisions.
- You are fluent with data and can collaborate effectively with data scientists and researchers.
- You have strong taste and real curiosity about model and agent behaviour.