AI Oral Exam Agents
Budget: $30 – $250 AUD
I want to build a pair of conversational AI agents that can take on the roles traditionally played by human examiners in an oral exam setting. The agents should be able to:
• Present questions aloud, wait for the student’s spoken response, and keep the flow of a typical exam conversation.
• Analyse the answer in real-time (speech-to-text plus basic natural-language understanding) and store a transcript.
• Hand off the transcript to an evaluation module that scores fluency, content, and pronunciation, then returns short, actionable feedback the student hears immediately.
• Log every exchange (audio + text + score) so I can later review performance trends.
At this stage I am open on the framework you choose—Whisper, DeepSpeech, Google Speech, Azure TTS, ElevenLabs, Rasa, LangChain, or even a custom PyTorch pipeline—as long as it runs reliably on a modest cloud instance and can be containerised for easy deployment.
Deliverables
1. A working prototype I can launch with Docker-Compose.
2. Simple web or desktop UI where I pick a question set, press “Start Exam”, speak, and see/hear the feedback.
3. Clean, well-commented code and a README that lets another developer extend it to other exam types or languages later.
Please outline your proposed tech stack, any pre-trained models you will leverage, and how many hours you expect for the first usable MVP.
• Present questions aloud, wait for the student’s spoken response, and keep the flow of a typical exam conversation.
• Analyse the answer in real-time (speech-to-text plus basic natural-language understanding) and store a transcript.
• Hand off the transcript to an evaluation module that scores fluency, content, and pronunciation, then returns short, actionable feedback the student hears immediately.
• Log every exchange (audio + text + score) so I can later review performance trends.
At this stage I am open on the framework you choose—Whisper, DeepSpeech, Google Speech, Azure TTS, ElevenLabs, Rasa, LangChain, or even a custom PyTorch pipeline—as long as it runs reliably on a modest cloud instance and can be containerised for easy deployment.
Deliverables
1. A working prototype I can launch with Docker-Compose.
2. Simple web or desktop UI where I pick a question set, press “Start Exam”, speak, and see/hear the feedback.
3. Clean, well-commented code and a README that lets another developer extend it to other exam types or languages later.
Please outline your proposed tech stack, any pre-trained models you will leverage, and how many hours you expect for the first usable MVP.