How vivekvt/mockmentor works
Project Type and Purpose
• Full-stack web application for AI-powered interview practice.
• Frontend: Next.js 15 + React + TypeScript + Tailwind.
• Backend: Next.js API Routes (serverless), MongoDB via Mongoose.
• External AI/ML services: OpenAI GPT, Google AI SDK, HeyGen streaming, Deepgram STT, ElevenLabs TTS.
• Authentication: Clerk.
• Real-time streaming: WebSockets (or SSE).
File Structure for Architectural Hints
• app/
• Pages & layout for “upload resume,” “new interview,” “ongoing interview” (React components in app/page.tsx, app/interview/[id]/page.tsx, etc.)
• serverless API endpoints in app/api/* for resume processing, job description processing, interview creation, access-token retrieval, WebSocket handshake, user profile.
• components/
• UI (Radix UI wrappers under components/ui)
• Domain logic hooks (components/logic/*) handling streaming, chat state, connection quality, etc.
• Avatar video component (HeyGen integration).
• lib/
• Configuration (appConfig.ts)
• Mongoose models & MongoDB connector (Interview.ts, User.ts, mongodb.ts)
• Utility functions (utils.ts).
• Public/static assets under public/.
• Root configs: next.config.ts, tsconfig.json, postcss.config.mjs.
Extract Key Architectural Insights from README
• Sequence: Resume upload → AI analysis → Job-description intake → Interview creation → Real-time voice/video stream → Feedback.
• Technologies for each step: resume (OpenAI), job analysis (OpenAI), video avatars (HeyGen), voice-chat (Deepgram + ElevenLabs), data storage (MongoDB), auth (Clerk).
a. Main Components
• Client Browser (React UI + hooks)
• Next.js Server (handles SSR/CSR & API Routes)
• Authentication Service (Clerk)
• Database (MongoDB via Mongoose)
• AI Services
• OpenAI GPT (resume & question generation)
• Google AI SDK (optional ML tasks)
• HeyGen Streaming Avatar API
• Deepgram STT (speech-to-text)
• ElevenLabs TTS (text-to-speech)
• WebSocket/Streaming Layer
b. Relationships & Data Flow
1. Client → Next.js Frontend (SSR/CSR pages)
2. Frontend ↔ Clerk for sign-in/out and session tokens
3. Frontend → API Route: upload-resume → Next.js → OpenAI (resume parse) → Store in MongoDB
4. Frontend → API Route: process-job → OpenAI (job-tailored questions)
5. User selects mentor → API Route: create-interview → Create Interview record in MongoDB
6. Client opens WebSocket to Next.js (or separate Node WS server) → negotiate streaming token via API Route: get-access-token → HeyGen + ElevenLabs + Deepgram
7. Live streaming data (audio/video chunks) flows through WS → Client AvatarVideo component renders stream
8. Conversation events stored/updated in MongoDB (message history) via API Routes or direct calls
c. Architectural Patterns & Principles
• Serverless/API-Route microservices pattern
• Separation of concerns: UI components vs. domain logic hooks vs. data layer (lib/)
• Event-driven streaming via WebSockets
• Single Source of Truth: MongoDB for interview sessions, user profiles, history
d. Technologies & Libraries
• Next.js App Router (app/ directory)
• React + TypeScript + Tailwind CSS
• Clerk for auth (middleware.ts)
• Mongoose & MongoDB (lib/mongodb.ts, lib/models/)
• HeyGen, OpenAI, Google AI SDK, Deepgram, ElevenLabs (lib/appwrite.ts, utils)
Essential Details
• Context provider (components/logic/context.tsx) linking UI → streaming hooks.
• Auth flow via Clerk (middleware.ts) before any API call.
• Real-time streaming separately from RESTful endpoints.