Voice AI is one of the fastest-moving model categories in 2026, powering clinical scribes, real-time customer service, transcription at scale. Like every AI product, it requires a core decision: rely on API's, or own the model. This workshop is for teams choosing ownership.

Any team buying third-party APIs inherits someone else's accuracy ceiling, their latency floor, and their pricing curve. API's may have earned their place in your stack but the teams getting serious about production are the ones running their own models and fine-tuning for their domain.

Join AWS and NVIDIA for a full-day, hands-on workshop: deploy NVIDIA Nemotron ASR on AWS infrastructure, fine-tune it for your domain and accents, and optimise inference for your latency budget.

Featuring: A real-world case study from Heidi Health, who will provide a firsthand account of their journey fine-tuning production-grade ASR for Australian clinical speech on AWS. What worked, what didn't, and what they'd do differently.

We use the voice stack as the teaching vehicle because it exercises every hard trade-off at once: latency, accuracy, streaming, and inference cost. The patterns transfer directly to LLM fine-tuning, embedding models, and multimodal workloads.  So if you're exploring fine tuning for non-voice use cases, you should come along too.

What You’ll Learn:  

• The full NVIDIA voice model stack — cascaded and unified architectures, and when to choose each  

• A repeatable fine-tuning evaluation loop (benchmark → adapt → validate) that works for any open-weight model, not just ASR  

• Deployment and runtime optimisation on Amazon EKS with G6 (L4 GPU) instances using Triton and ONNX  

• How to reason about the cost, latency, and accuracy trade-offs of self-hosting vs. API

What You’ll Leave With: 

• A mental model of model ownership — architecture selection through to runtime optimisation — demonstrated through voice but directly transferable to LLM and embedding workloads  
• A production-ready ASR deployment on AWS Containers  
• Confidence to apply the same deploy → tune → optimise → scale pattern to TTS, LLM, or multimodal serving