Job Description & Details
This is a hands-on Applied ML role focused heavily on shipping real models that actually power user-facing products, rather than just doing endless R&D in a notebook. You'll own the entire lifecycle here, from building out retrieval and ranking systems to monitoring inference latency and cost in production.
What You'll Actually Be Doing
Your typical week involves a mix of model iteration, pipeline tuning, and performance debugging. You might spend Monday optimizing an embedding-based search retrieval pipeline to shave off a few milliseconds of latency, and Tuesday setting up evaluation metrics for a new personalization feature. The core challenge is balancing raw model accuracy with practical constraints like inference cost and real-world system reliability.
The Core Tech Stack
You need to be fluent in Python and comfortable building production systems using PyTorch or JAX. The engineering team relies heavily on vector search, modern neural architectures, and robust inference pipelines. Knowing how to tune models for ranking, classification, and hybrid search while keeping an eye on resource utilization is non-negotiable for keeping these AI-native features running smoothly.
Interview Expectations
Expect the team to drill down into your experience taking models from local prototypes all the way to production deployment. They will likely ask you to debug a hypothetical production scenario where model latency spiked or retrieval relevance degraded after a data drift. The interviewer isn't just looking for a textbook algorithm answer; they want to see your pragmatic engineering trade-offs between model complexity, infrastructure cost, and user experience.
Application Advice
Make sure your resume doesn't just list theoretical machine learning concepts, but explicitly highlights the production systems you have built and deployed. Sprinkle in keywords from the job description like PyTorch, vector search, classification, ranking models, and inference pipelines. Emphasize metrics-driven results—tell them how your model improvements tangibly impacted latency, precision, or business value so you breeze right past the automated filters.