GIF
Website Gif

About MockMentor

💡 What Inspired This Project

The inspiration for MockMentor came from a personal pain point I've experienced throughout my career journey. As someone who's been through countless interviews, I noticed that the biggest barrier to interview success isn't knowledge—it's confidence.

Traditional mock interviews require scheduling with friends or mentors, generic practice questions don't reflect your unique background, and many people (myself included) felt awkward practicing with other humans. I wanted to create a solution that would:

Be available 24/7 without scheduling conflicts
Provide personalized questions based on your actual resume
Feel realistic enough to build genuine confidence
Remove the social anxiety of practicing with real people

When I discovered HeyGen's streaming avatar technology, I realized I could create something truly revolutionary—AI mentors that look and feel like real interviewers.

🚀 How I Built It

Tech Stack & Architecture

Frontend: Next.js 15 + React 19 + TypeScript + Tailwind CSS
AI/ML: Google AI SDK, OpenAI GPT, HeyGen Streaming Avatar API
Voice: ElevenLabs voice models + Deepgram speech-to-text
Backend: Next.js API routes + MongoDB + Mongoose
Auth: Clerk for seamless user management

Development Process

Phase 1: Core Infrastructure (Day 1)

// Set up the streaming avatar session management
const { initAvatar, startAvatar, sessionState, stream } = useStreamingAvatarSession();

// Configure real-time voice chat
const config: StartAvatarRequest = {
  quality: AvatarQuality.High,
  voiceChatTransport: VoiceChatTransport.WEBSOCKET,
  sttSettings: { provider: STTProvider.DEEPGRAM }
};

Phase 2: AI Integration (Day 2)

Implemented resume parsing and analysis using Google AI
Built job description processing for tailored questions
Created intelligent conversation flows that adapt to user responses

Phase 3: User Experience (Day 3)

Designed responsive UI with real-time video streaming
Added timer functionality for focused 3-minute sessions
Implemented mentor selection with 6 different AI personalities

🧠 What I Learned

Technical Insights

Real-time Streaming: Managing WebSocket connections while maintaining smooth video performance requires careful state management and error handling.
AI Orchestration: Combining multiple AI services (GPT for conversation, ElevenLabs for voice, Deepgram for STT) taught me about API latency optimization and fallback strategies.
Context Management: Building AI that remembers both resume details AND job requirements while maintaining natural conversation flow was more complex than expected.

Product Insights

User Experience: The difference between "technically working" and "feels natural" is enormous—small details like avatar lip-sync and voice emotion make huge impact.
Performance Matters: Users expect instant responses in conversational AI. Every 100ms delay breaks immersion.
Progressive Enhancement: Starting with core functionality (text chat) then adding voice made development more manageable.

🚧 Challenges Faced & Solutions

Challenge 1: Streaming Performance

Problem: Video streaming would lag when processing voice input simultaneously.

Solution: Implemented efficient state management and optimized WebSocket connections:

// Optimized avatar event handling
avatar.on(StreamingEvents.AVATAR_START_TALKING, (e) => {
  console.log('Avatar started talking', e);
});
avatar.on(StreamingEvents.STREAM_READY, (event) => {
  setLoading(false);
  setStartTime(new Date());
});

Challenge 2: Context-Aware AI Conversations

Problem: Making the AI understand both resume content AND job requirements while maintaining natural flow.

Solution: Developed a sophisticated prompt engineering system that:

Analyzes resume for key skills and experience
Extracts job requirements and responsibilities
Generates contextual questions that bridge both
Maintains conversation memory throughout the session

Challenge 3: Natural Voice Interactions

Problem: Achieving realistic back-and-forth conversation without awkward pauses or overlaps.

Solution: Integrated multiple services with careful timing:

Deepgram: Real-time speech-to-text with low latency
ElevenLabs: Natural voice synthesis with emotion
Custom logic: Interrupt handling and turn-taking management

Challenge 4: Mobile Responsiveness

Problem: Streaming video and voice chat on mobile devices with varying network conditions.

Solution:

Implemented adaptive quality settings
Added fallback options for slower connections
Optimized UI for touch interactions and smaller screens

🎯 Key Features Implemented

6 AI Mentor Personalities: Each with unique conversation styles
Smart Resume Analysis: AI extracts and understands your background
Job-Tailored Questions: Custom interviews based on job descriptions
Real-time Voice Chat: Natural conversations with interrupt handling
Timed Sessions: Focused 3-minute practice rounds
Cross-platform: Works seamlessly on desktop and mobile

🔮 What's Next

The hackathon version proved the concept works beautifully. Next steps include:

Performance Analytics: Detailed feedback on speaking pace, filler words, confidence
Industry Specialization: Tailored interview modules for tech, finance, healthcare, etc.
Team Interviews: Multi-person interview simulations
Integration: Connect with job boards for real-time practice opportunities

🏆 Impact & Reflection

Building MockMentor in 3 days taught me that the best products solve real problems you've experienced yourself. The combination of cutting-edge AI technology with genuine user need created something that feels magical to use.

The most rewarding moment was testing the first complete interview session—having a natural conversation with an AI mentor about my actual experience felt surreal and genuinely helpful for building confidence.

This project showcases how modern AI tools can create entirely new categories of solutions that weren't possible just a year ago. We're living in an incredible time for builders.