An AI mock interview product for HKUST major selection interviews. It helps students practice with personalized questions, multi-round follow-ups, voice simulation, and structured feedback, while validating campus interview coaching demand and accumulating early users for a future general interview product.
Project Background
Students preparing for major selection interviews often face three key problems:
- Limited access to real question banks and unclear assessment focuses across majors
- Lack of in-depth practice based on personal experience, as traditional question banks are mostly generic
- Lack of structured feedback on expression, logic, motivation, and improvement areas
General voice interview models cannot fully cover past question patterns, major-specific differences, or personalized follow-up needs. Offer Plus was built as a campus-focused entry point to validate real demand for AI interview practice.
Project Highlights
- Personalized questions: Generates questions based on CV / PS, target major, and past question banks
- Multi-round follow-ups: Probes project details, motivation, reflection, and supporting evidence based on user answers
- Four-dimensional assessment: Provides structured feedback on expression, logic, major fit, and reflection depth
Reached 2,000+ visits within 7 days of launch, covering over 30% of the target interview population;
Users completed 200+ mock interviews;
Nearly 80% of users were satisfied with question depth and overall experience.
System Architecture
- Frontend and backend: Built with Next.js
- Identity and data: Supabase for login, user data, interview records, reports, and question banks
- AI generation: Gemini text models for personalized questions, follow-ups, and structured reports
- Real-time voice: Bridge service connects the frontend with Vertex Live API for voice Q&A, transcription, and interview state control
Key Optimizations
First-question waiting time
Based on user interviews and funnel data, we moved question generation earlier to the onboarding / interview creation stage, reducing first-question waiting time by around 80%.
Follow-up stability
To avoid unstable or unexpected voice output, we separated text generation from speech: the text model first generates a JSON follow-up, the backend validates it, and the Live model then reads it aloud.
Full Feature Set
Student Side
- Onboarding: Select target majors and upload or fill in CV / PS
- AI mock interview: Voice-based Q&A with multi-round follow-ups
- Structured report: Overall score, dimension-level feedback, question-level suggestions, and improvement directions
- Retake practice: Practice again with the same interview plan
Operations and Management Side
- Question bank management: Maintain past question banks for different majors
- User and interview management: Track completion, report status, and abnormal cases
- Feedback collection: Collect feedback on question depth, interview experience, and report quality