ML systems engineer focused on inference, GPU kernels, and compilers.

Built production audio ML systems across music search, stem separation, and on-device inference alongside Spotify's former Head of Innovation. Created an end-to-end ETL pipeline for 1M+ songs, optimized BS-RoFormer to 9.65 average SDR on MUSDB18 with sparse attention and quantization, and built agentic workflows that reduced manual intervention by 60% while accelerating tasks by 2.5x.

Co-authored research on ConstraintFlow DSL compiler optimization for neural network verification, refactored core IR passes, and created 15+ technical diagrams. Engineered multi-level caching and JIT compilation for a 51x critical-path speedup, designed a CPU-GPU synchronization and scaling roadmap, and reduced end-to-end compile time by 37% through profiling across 20+ passes.

Relevant coursework: Data Structures & Algorithms, Operating Systems, Systems Programming, Distributed Systems, Advanced Computer Architecture, Linear Algebra, Parallel Programming, ML Systems, and Generative AI.