Diolex

Stop grinding LeetCode. Start training like it's real. Diolex trains the real skill: navigating unknown problems through effective communication, not memorization.

Comment

Diolex: The Conversational AI Interview Simulator

The Problem: Traditional interview prep focuses on pattern recognition, not the crucial "meta-skills" of communication, strategic questioning, and hint extraction vital for real technical interviews.

Our Solution: Diolex is a voice-first AI-powered simulator designed to train these essential meta-skills. Our AI interviewer:

  • Watches code in real-time: Provides contextual feedback based on your approach.
  • Teaches strategic questioning: Withholds information, prompting you to ask clarifying questions.
  • Simulates authentic dynamics: Engages in follow-up questions, hint extraction, and edge-case discussions.
  • Provides detailed analysis: Offers specific feedback on communication and problem-solving.

How We Built It

Frontend:

  • React + TypeScript: For robust, type-safe components.
  • Tailwind CSS: For rapid, responsive styling.
  • CodeMirror 6: Provides a syntax-highlighted code editor.
  • Custom WebSocket Hooks & Speech Recognition API: Enables real-time, bidirectional communication and continuous voice input.
  • React Router: Manages seamless navigation.

Backend:

  • FastAPI: A high-performance asynchronous API.
  • WebSockets: For real-time voice and text communication.
  • Custom Interview Agent: A structured, stateful AI with authentic interviewer persona.
  • Kokoro TTS: For natural-sounding spoken feedback.
  • Piston API: Secure sandboxed code execution.
  • SQLAlchemy + PostgreSQL: For reliable data persistence.

AI & Voice Technology:

  • Finetuned Outputs & Context-Aware Responses: Ensures authentic, adaptive conversations based on code and history.
  • Multi-modal Interaction: Supports both voice and text.
  • Intelligent Hint Distribution: Provides strategic guidance without giving away answers.

Key Technical Challenges Overcome

  1. Real-time Voice + Code Synchronization: We built a sophisticated WebSocket message queue system with prioritization and conflict resolution to ensure seamless integration of speech recognition, code editing, and AI responses.
  2. Context-Aware AI Responses: Developed a dynamic context injection system that sends code snapshots with every message, allowing the AI to intelligently reference your live implementation.
  3. Authentic Interview Simulation: Achieved realistic AI behavior through extensive prompt engineering, multi-phase interview logic, information withholding strategies, and natural conversation flow patterns.
  4. Cross-browser Speech Recognition: Implemented robust fallback mechanisms, automatic restart logic, and graceful degradation to text-only mode to counter browser inconsistencies.
  5. Low-latency Voice Responses: Streamed TTS with chunk-based audio playback and WebSocket message prioritization to minimize delay for natural conversation flow.

What We Learned

Technical: Mastered WebSocket architecture, advanced speech API integration, AI prompt engineering for conversational AI, React performance optimization, and FastAPI async patterns.

Product: Understood the critical impact of authentic simulation, unique UX considerations for voice interfaces, and the importance of seamless transitions for users.

Startup: Validated our core hypothesis, discovered new use cases (e.g., explaining solutions), and recognized the scalability potential for different interview styles.

Future Vision

Diolex proves the power of AI to authentically simulate complex human interactions. Our vision is a comprehensive interview preparation platform that adapts to diverse company styles, skill levels, and formats, revolutionizing career readiness for developers.

Built With

Share this project:

Updates