AI interview practice agent

Aether

Aether is an interview practice environment designed around live interaction instead of post-hoc grading. The point is to keep the conversation responsive enough that users feel coached in the moment, then leave with feedback that is specific enough to act on.

Open app

The problem it tackles

Most interview tooling either records a session and grades it later or wraps a chat UI around canned prompts. Aether had to feel closer to an active interviewer: low-latency audio handling, structured feedback, and a conversation loop that could stay coherent while the system was still reasoning in the background.

Architecture

The key system decisions

Streaming interview loop

Audio is ingested continuously so the product behaves like a live interviewer instead of a submission portal waiting on a final transcript.

Session control and auth

JWT-based session boundaries keep practice runs isolated while a compact Go backend holds onto the state needed for fast turn-taking.

Feedback synthesis

The system translates raw conversation into actionable signals so users get critique on clarity, structure, and delivery rather than vague scoring.

What matters

Designed the interaction flow around speaking, reflecting, and iterating instead of filling a form and waiting.
Kept the backend small enough to stay understandable while still supporting real-time behavior.
Tightening the feedback model so coaching feels specific rather than generically supportive.

GoDeepgramOpenAIWebSocketsSQLiteJWT

Media

Screenshots and demos

Aether live conversation screen

Aether demo video