InterviewAce

• 

  InterviewAce helps candidates practice smarter, improve faster, and build confidence with AI-powered interview coaching.

• 

  Users can customize role, difficulty, interview type, language, and question source for personalized practice.

• 

  Each answer receives instant AI feedback with scoring, strengths, weaknesses, and likely follow-up questions.

• 

  InterviewAce suggests better answer examples to help users learn how to respond more clearly and effectively.

• 

  Every session ends with a full report including overall score, readiness summary, weak areas, and next steps.

• 

  Practice history lets users revisit past sessions, resume incomplete interviews, and track improvement over time.

• 

  The user workspace includes preferences, account controls, data export, and quick access to practice tools.

Inspiration

Many candidates read question lists but rarely practice answering them under realistic conditions. I wanted a single product that simulates real interviews (voice + text), gives structured AI feedback, and helps users measure true progress over time. That became InterviewAce.

What it does

InterviewAce enables:

• Voice and text mock interviews with instant AI scoring and actionable feedback
  
• Resume-driven and JD-driven question generation for role-specific practice
  
• Bilingual support and on-demand report translation
  
• Context-aware questions using news and web search
  
• Progress tracking, history, and data export

How I built it

I built InterviewAce as a React + TypeScript frontend with Supabase for auth, storage, and DB. Core AI & multimodal capabilities are exposed via Edge Functions that call external plugins/APIs. Key components:

• VoiceRecorder & AudioPlayer for TTS/STT flows
  
• ResumeUpload & JDImport for resume/JD extraction and personalization
  
• generate-interview Edge Function enhanced to accept context: resume_text, jd_extracted, news_context, search_context

Data model changes include storing session-level flags (voice_mode, resume_based, translation_enabled) and a voice_recordings table. Average session score is computed as usual, e.g.: [ \bar{S} = \frac{1}{n}\sum_{i=1}^{n} S_i ] where (S_i) are completed session scores.

Challenges

• Making AI feedback feel actionable (structuring scores, strengths, and improvement steps)
  
• Reliable speech transcription across Chinese/English with low latency and cost controls
  
• Extracting JD content robustly across varied job sites and cleaning HTML noise
  
• Ensuring privacy and data isolation (RLS, storage policies) while persisting audio artifacts

What I learned

• AI UX requires more than model outputs — product flow, persistence, and clear explanations matter.
  
• Multimodal features increase complexity (privacy, storage, cost) but greatly improve realism and user engagement.
  
• Edge Functions are effective for isolating third-party integrations and enforcing security.

Next steps

Refine personalization (saved practice plans), extend analytics (mode effectiveness), and add admin tools for curriculum & monitoring.

Built With

• medo
• postcss
• postgresql
• react
• supabase
• tailwind-css
• typescript
• vite