Efficient Speech Serving

Reduce the cost of inference

I worked on a Rust speech inference service to reduce the overhead of an existing serving approach. The task covered the inference path and the application behavior around it.

I focused on efficient execution, bounded resource use, and a predictable interface. Performance comparisons helped assess the tradeoffs in speed and memory rather than assuming a language change would solve them.

The project brought systems programming and model serving together. It reinforced the value of measuring the complete service, including work outside the model itself.