How prish20/prepwise-interview works
Project Type and Purpose
• This is a full-stack web application: an AI-powered mock interview platform (“Prepwise”) built with Next.js (App Router) on the frontend and Firebase (Auth + Firestore) on the backend.
• It leverages external AI services (Vapi for real-time voice interviews, Google Gemini for question generation, OpenAI GPT-4 for feedback).
File Structure
• Top-level:
• app/ (Next.js pages and layouts, segmented into (auth) and (root) for public vs. authenticated routes)
• app/api/vapi/generate/route.ts (API route for question generation)
• components/ & components/ui/ (reusable UI components with ShadCN styling)
• firebase/ (client.ts and admin.ts for Firebase SDK setup)
• lib/ (business logic and SDK wrappers: auth.action.ts, general.actions.ts, vapi.sdk.ts)
• constants/ (static configuration)
• types/ (TypeScript type definitions)
• Patterns:
• App Router – layout.tsx files define global and nested layouts
• Feature grouping – (auth) and (root) route groups
• API route encapsulating AI workflows
README for Additional Insights
• Architecture overview in “How It Works”:
1) Firebase Auth with session cookies → secure protected routes
2) POST to /api/vapi/generate → Gemini question generation → save to Firestore
3) Real-time voice interview via Vapi SDK
4) Transcript saved → GPT-4 feedback analysis → detailed scoring
• Tech stack clearly calls out Next.js, Firebase, Vapi, Gemini, OpenAI, Tailwind CSS, ShadCN UI, React Hook Form + Zod.
Main Components
a. Frontend (Next.js App Router)
• Layouts & pages: sign-in, sign-up, interview list, interview session
b. Authentication Layer
• Firebase Auth (client SDK)
• Session cookie management (admin SDK)
c. API Layer (Next.js API Routes)
• /api/vapi/generate
• Business logic (lib/actions, lib/vapi.sdk.ts)
d. Data Storage
• Firestore (interview records, transcripts, feedback)
e. External AI Services
• Vapi SDK (real-time voice interaction)
• Google Gemini API (question generation)
• OpenAI GPT-4 (feedback analysis)
f. Static Assets & Constants
• Public assets, constants for tech stacks, prompts
Relationships and Interactions
• User (browser) → Next.js Frontend → Firebase Auth → Session Cookie
• Authenticated Frontend → Interview UI → calls API route (/api/vapi/generate)
• API route → Gemini API (questions) → Firestore → returns questions to UI
• Interview UI ↔ Vapi SDK ↔ Vapi service (streaming voice)
• On call end: UI sends transcript → API route or lib/sdk → OpenAI GPT-4 → Firestore (feedback) → UI fetches/display feedback
Architectural Patterns and Design Principles
• Separation of concerns: presentation (app/pages + components), business logic (lib/actions), data layer (Firebase SDK, Firestore), external integrations (Vapi, Gemini, GPT-4)
• Feature-based routing with Next.js App Router
• Secure server-side session management via Firebase Admin