Interview Prep Coach

Inspiration

We created Interview Prep Coach after watching talented peers—especially from underrepresented backgrounds—fail behavioral interviews despite strong technical skills. They struggled to articulate experiences using the STAR method. Professional coaching was prohibitively expensive ($100-300/session), and generic practice tools offered no structured feedback. We saw an opportunity to democratize interview preparation through responsible AI.

What it does

Our platform transforms behavioral interview practice with:

AI STAR Analysis: Real-time feedback on Situation, Task, Action, Result components with 0-100 scoring

Dual Input Modes: Voice recording with silence detection or text input

Resume-Powered Questions: AI parses resumes to generate personalized behavioral questions based on actual experiences

Customizable Coaching: Users control feedback emphasis, tone (Encouraging to Strict), and detail level

Progress Analytics: Visual dashboards track trends and identify improvement patterns

Educational Focus: Provides "Improved Version" rewrites showing exactly how to strengthen answers

Built with bias mitigation, transparent scoring criteria, and privacy-first design.

Challenges we ran into

Full-Stack Integration: Orchestrating four layers (frontend, backend, database, AI) required strict API contracts and end-to-end type safety with Prisma.

Database Schema Evolution: Our simple Q&A design pivoted to support sessions, resume parsing, and granular feedback. Prisma migrations helped us iteratively refactor relationships mid-hackathon.

AI Reliability: Gemini's free-form responses broke our JSON pipeline. We implemented defensive parsing with json_repair and engineered prompts with explicit schemas

Audio Processing: Real-time silence detection failed across devices. We calibrated Web Audio API thresholds dynamically based on ambient noise.

Accomplishments that we're proud of

Responsible AI: Transparent scoring, customizable feedback, and bias mitigation prove AI can be ethical and powerful.

End-to-End System: Shipped production-ready authentication, resume parsing, audio recording, AI feedback, and analytics.

What we learned

Next.js + Express Decoupling: Mastered handling CORS, auth, and deployment complexities of separate frontend/backend services

Prompt Engineering: AI reliability depends more on prompt structure than model power—explicit JSON schemas and bias instructions dramatically improved outputs

Schema-First Design: Late-stage database changes cascade through the entire stack; whiteboarding relationships early saves hours