TrialHaus: AI interview practice

Context

Interview practice needed a conversation that could respond to previous answers, rather than a fixed list of questions. The product combined video and voice capture with role-specific questions and feedback. Session context needed to persist as the interview progressed.

System and approach

WebRTC handles the session media, and Whisper turns speech into text. An LLM generates questions using the target role, seniority and prior answers. Transcript and answer content support feedback, while session state allows subsequent questions to build on the conversation.

• Capture sessions through WebRTC and transcribe speech with Whisper.
• Generate questions from role, seniority, and prior answers.
• Use transcript and answer content to shape feedback.
• Keep session state across turns for coherent follow-ups.

Delivered scope

The engineering scope includes real-time media, transcription, adaptive questions and interview feedback within a React application.

Technology

• Python
• FastAPI
• OpenAI
• WebRTC
• Whisper
• React