TinySSM: Keyword Spotting at the Edge
Exploring state-space models for always-on keyword spotting within the memory and latency limits of microcontrollers.
- TinyML
- Keyword spotting
- State-space models
- Edge AI
Projects & directions
Efficient intelligence is a systems problem. I am interested in how models, kernels, and agents can be designed together so that useful AI runs within real constraints.
Exploring state-space models for always-on keyword spotting within the memory and latency limits of microcontrollers.
Investigating whether language-model agents can help write, test, and refine the low-level kernels that make inference practical on microcontrollers.
The projects are specific, but the questions overlap. These are the directions that connect them.
Models and training methods that deliver useful capability within real compute, memory, and energy budgets.
Language-model agents that can inspect a problem, use tools, test ideas, and improve software systems.
Always-on sensing and inference on microcontrollers and other devices at the far edge of the network.
Kernels, memory layout, quantization, and deployment choices that make inference faster and more predictable.
No peer-reviewed publications are listed yet.
The project pages are the most current record of the work. Published papers and preprints will be listed on the publications page when they exist.