Senior Conversational AI Engineer – Privacy-First ML

Apple, Inc.; Seattle (WA); On-site; USD 175,000 - 309,000; Full time

Benefits offered by this job

Stock awards

Employee stock purchase plan

Education reimbursement

Relocation assistance

Job summary

Apple is seeking a Senior Machine Learning Engineer in Seattle to advance offline evaluation insights for Siri, focusing on privacy-preserving, personalized conversational AI. You will work within the Siri organization, contributing to Apple Intelligence and SCALE, shaping evaluation at scale across iOS, iPadOS, macOS, watchOS, and visionOS.

Collaborating with cross-functional teams, you will build and curate high-quality evaluation datasets for state-of-the-art models, develop adversarial

Qualifications

• 7+ years of ML experience applying ML to real-world problems, especially in natural language products.
• Experience managing large-scale datasets for ML training and evaluation.
• Excellent programming skills in Python.
• MS/PhD in Machine Learning, Computer Science, or equivalent experience in a related field.

Responsibilities

• Leverage Large Language Models (LLMs) to automatically evaluate model changes (LLM-as-a-Judge) and assess the naturalness of digital assistant conversations.
• Harness generative AI to create adversarial scenarios that anticipate edge cases, enabling robust evaluation.
• Contribute to shaping the next generation of Siri and Apple products through dataset generation and evaluation.

Skills

Python

Machine Learning

Large-scale data handling

Education

MS/PhD in ML or CS

Job description

Apple is seeking a Senior Machine Learning Engineer in Seattle to advance offline evaluation insights for Siri, focusing on privacy-preserving, personalized conversational AI. You will work within the Siri organization, contributing to Apple Intelligence and SCALE, shaping evaluation at scale across iOS, iPadOS, macOS, watchOS, and visionOS.

Collaborating with cross-functional teams, you will build and curate high-quality evaluation datasets for state-of-the-art models, develop adversarial