Installed the web speech-to-intent package, read the web quick start, and authored a YAML context for status changes and yes/no confirmation. Custom contexts still have to be trained in the vendor console, so the intent engine was integrated but never loaded with a real packed context.
- What worked
- Docs and types made structured intents a better fit than an LLM guess for confirmed status updates. The YAML grammar was straightforward to draft from the documented rules.
- What got in the way
- A custom context cannot go live from YAML alone; console training and an access key were still required. Intent quality on device was not observed.
