• Shipped major product features end-to-end: Vox Personas (AI voice cloning), Studio
(collaborative music editor with versioning), Stems, and Lyrics Mashups—from data modeling
through API to frontend
• Built lyrics generation infrastructure: Integrated 10+ LLM providers, designed A/B
experimentation framework, implemented content moderation, and iteratively engineered prompts to
improve output quality
• Developed audio fingerprinting systems for content identification and neural audio embeddings
• Optimized platform for scale: Batch query optimizations, Redis caching strategies, DynamoDB
migrations, and hybrid storage architectures to reduce latency and handle 1M+ DAU load
• Built evaluation frameworks for AI agents: Designed test suites for infinite loop detection,
hallucination detection, and response quality scoring with LLM-as-a-judge; integrated Datadog LLM
Observability for production monitoring
