Responsibilities
- Train and fine-tune ASR models at scale, including dataset strategy, augmentation, and domain adaptation to real-world clinical audio.
- Build and improve validation and evaluation frameworks, including WER and targeted analysis across speakers, environments, devices, and clinical terminology.
- Deploy and operate ASR inference services with focus on reliability, latency, and efficiency in production.
- Optimize inference latency and throughput, including batching strategies, model export choices, and hardware-aware profiling.
- Build and maintain APIs and services in frameworks like FastAPI, Kafka, and NVIDIA Triton, and deploy and run them on Kubernetes.
- Take technical ownership of core ASR components, shaping best practices for modelling, evaluation, and production reliability across the team supporting the growth of engineers working on speech systems.
- Work closely with product and platform teams on safe rollouts, monitoring, and continuous improvement based on real-world feedback.
Requirements
- Strong programming skills in Python and the ability to contribute to production-grade codebases.
- Hands-on experience in speech recognition and ASR, including at least some of the following: Training or finetuning end-to-end ASR models (CTC/TDT/AED) at meaningful scale.
- Designing evaluation and validation approaches, including WER and deeper error analysis.
- Deploying and operating ASR services in production, including performance and reliability work.
- Experience building ML systems that can be deployed and operated, including pipelines, CI and CD practices, and monitoring.
- Clear communication and collaboration skills across research, engineering, and product.
- A Master’s degree in computer science, engineering, mathematics, statistics, physics, or a related field, or equivalent professional experience.
Nice to Have
- Experience with multilingual ASR, streaming inference, noisy audio conditions, or healthcare privacy and compliance constraints.
- Experience with Go.
Team
Reports to: Director of Engineering
Additional Information
- The position is full-time and starts as soon as possible
- Equipment provided by Corti
