The problem InquisAI solves

Problem It Solves

• For Students:

  • Provides a scalable, consistent, and detailed solution for technical interview preparation.
  • More effective and accessible compared to traditional methods like peer mock interviews or self-practice.
  • Helps users practice a wide range of topics, including:
    • Frontend
    • Backend
    • Cloud
    • Mobile
    • Programming
  • Includes relevant tech stacks and tools.

• For Companies:

  • Efficiently manages interviews for 1000+ applicants for the same job listing.
  • Automates the initial interview stage with a realistic 3D model interviewer.
  • Conducts numerous mock interviews simultaneously without requiring human interviewers.
  • Significantly reduces time and resource expenditure.
  • Records and analyzes candidate responses.
  • Provides detailed feedback on:
    • Clarity
    • Accuracy
    • Completeness
    • Relevance
    • Communication
  • Ensures only the most qualified candidates proceed to further interview rounds.
  • Streamlines the hiring process and improves overall efficiency.

Challenges we ran into

Challenges We Ran Into

Realistic 3D Model Integration

Integrating a realistic and interactive 3D model interviewer using three.js while ensuring smooth performance and synchronization with audio files posed a significant challenge. We addressed this by using optimized open-source 3D models to maintain a seamless experience.

Audio and Lipsync Generation

Generating accurate lipsync data and audio files dynamically using AWS Polly was challenging. To overcome this, we implemented a caching mechanism to store and reuse previously generated files, reducing latency and resource usage.

Real-Time WebSocket Management

Ensuring that each user session is unique and secure, preventing multiple concurrent sessions, was difficult. We established a robust WebSocket connection with server-side checks to enforce single-session rules.

Automatic Recording and Pause Detection

Implementing an automatic recording system that accurately detects pauses and stops recording appropriately was challenging. We utilized precise audio analysis algorithms to detect pauses exceeding 4 seconds and trigger automatic stop.

Speech-to-Text Accuracy

Achieving high accuracy in transcribing user responses using the Deepgram API required extensive testing and fine-tuning of the API settings to improve transcription accuracy.

Feedback and Evaluation System

Developing a comprehensive feedback system using OpenAI's API that evaluates responses on multiple criteria was challenging. We performed thorough prompt engineering to refine the evaluation prompts and ensure reliable and relevant feedback.

Scalability and Performance

Ensuring the system can handle multiple users simultaneously without performance degradation was a significant challenge. We deployed the backend on AWS EC2 for scalability and used efficient database queries to handle high loads.