Fullstack developer · Aury · AI counselling
Voice check-ins for support between sessions
What a check-in is
Aury is a stress-management course developed with psychologists and therapists in Germany. Between course sessions, people sometimes need help right away. A check-in lets them open the app and talk through what is on their mind with an AI counsellor, out loud, and get a spoken reply. It is confidential and anonymous.
What made it hard
Someone who is stressed won't press a button, wait, and talk in turns. The conversation has to flow like a phone call. So audio streams in both directions, and turn detection works out when the person has actually finished speaking before the AI answers. And because people bring real problems, every response has to be safe, which is why the testing gets as much attention as the pipeline.
Person checking in
Talks freely, can pause
LiveKit
Streams audio, detects when they've finished
Speech to text
Streams as the person talks
Counsellor agent
LangChain, models on Vertex AI
Text to speech
Streams the reply back
- Agents test agents. QA agents run whole conversations, instead of checking fixed inputs against fixed outputs.
- Partners simulated. Mocked integrations, conversations and recommendations, so the platform could be validated without the partner's live systems.
- LLM-assisted data work. An auto-tagging script labels large datasets, which feeds recommendation algorithms tested against cases taken from user stories.




