I work on applied AI systems with an emphasis on deployable machine learning, real-time perception, and resource-aware model design. My projects focus on building end-to-end ML pipelines that remain robust under practical constraints such as limited compute, latency requirements, and noisy real-world data.

• Embedded & Resource-Efficient Machine Learning
  Designing and deploying models for constrained environments, with attention to model size, inference latency, and performance trade-offs. Experience includes model compression, quantization, and edge-oriented deployment workflows.

• Real-Time Audio & Time-Series Perception
  Building streaming inference systems that combine signal processing and learning-based models for low-latency classification and detection. Emphasis on end-to-end evaluation, from preprocessing pipelines to system-level performance metrics.

• Model Adaptation & Learning Systems
  Exploring efficient strategies for adapting models when data or compute is limited, including parameter-efficient fine-tuning approaches and empirical comparison against full retraining baselines.

• Reinforcement Learning & Representation Analysis
  Experimental reinforcement learning systems focused on policy learning dynamics, representation choices, and stability, with structured evaluation and reproducible experimentation.

PyTorch, TensorFlow, parameter-efficient adaptation methods, audio signal processing, TensorFlow Lite, quantization and profiling, reinforcement learning frameworks, Python, C/C++, modular and reproducible ML pipelines.