Machine Learning Engineer | High-Performance Systems | C++/CUDA | Computer Vision | ML Infrastructure

I build real-time, performance-critical systems at the intersection of computer vision, GPU acceleration, robotics simulation, and ML infrastructure.
My background spans AI/ML, optimization, deep learning, GPU programming, and systems-level engineering, with a Master’s degree in AI & Machine Learning from Columbia University.

I'm currently focused on:

• Modern C++ (C++17/20) for vision, tracking, and real-time pipelines
• CUDA kernel optimization and GPU acceleration
• Inference runtime systems (batching, streaming, KV-cache simulation)
• Orchestrated ML pipelines (Airflow/Prefect, MLflow, Ray)
• Distributed GPU workloads on Kubernetes
• Simulation and autonomous robotics

• C++17/20, Python
• Linux, CMake, gdb, perf
• Multithreading, lock-free queues
• Real-time computer vision (OpenCV, tracking)

• CUDA, Nsight Systems / Nsight Compute
• TensorRT, ONNX Runtime
• Triton Inference Server
• GPU memory optimization, fused kernels

• PyTorch, TensorFlow
• FastAPI / gRPC
• Continuous batching, streaming inference
• Quantization (bitsandbytes), ONNX export
• Model deployment & benchmarking

• MLflow, Airflow / Prefect
• Docker, Kubernetes, Ray
• AWS / GCP
• Feature engineering & data validation

A high-performance, multi-threaded pipeline for real-time object tracking across multiple video streams.
Demonstrates C++ systems design, threading, frame pipelines, low-latency processing, and modular CV architecture.

Tech: C++17, OpenCV, pthreads, lock-free queues, CMake, Linux

Hand-optimized fused CUDA kernel implementing RGB→Gray, Gaussian blur, and Sobel edge detection in a single pass.
Includes CPU vs GPU benchmarks and profiling analysis.

Tech: CUDA, Nsight Systems/Compute, memory coalescing, shared memory

Custom inference engine with simplified continuous batching, token streaming, and latency profiling.
Builds intuition for LLM runtime engines (vLLM, SGLang, TensorRT-LLM).

Tech: FastAPI, PyTorch, queuing, GPU inference, streaming APIs

End-to-end ML workflow with orchestration, metrics tracking, model registry, CI/CD, and Dockerized deployment.

Tech: Airflow/Prefect, MLflow, S3, Docker, GitHub Actions

Lightweight GPU workload scheduler distributing inference jobs across a GPU-enabled Kubernetes cluster.
Includes autoscaling, monitoring, and queue-based scheduling.

Tech: Kubernetes, Triton, Prometheus, Grafana, Ray, Python

• Building deeper expertise in:
  • C++20 concurrency
  • CUDA warp-level optimization
  • LLM fault-tolerant runtime systems
  • Robotics simulation and GPU-accelerated perception
• Publishing a series of GPU + C++ systems projects over the next 8 weeks
• Preparing for ML Systems & Robotics Engineering roles

• LinkedIn: https://www.linkedin.com/in/drew-kalasky/ Open to opportunities in: ML Systems, GPU optimization, robotics simulation, inference infrastructure, applied ML engineering

1. Real-Time Multi-Camera Tracker (C++17)