TrueView

Conduct a fair interview: smarter, not harder

Comment

Inspiration

There’s a growing need to modernize how companies evaluate talent. Traditional interviews often rely on human perception, which can unconsciously skew decisions and overlook real capability. Research shows that 42% of women have faced biased or inappropriate interview questions. Studies from Harvard and the National Bureau of Economic Research reveal that applicants with “ethnic-sounding” names receive up to 50% fewer callbacks, and candidates from lower-income or non-Ivy League schools face systemic barriers despite equal qualifications. Bias extends far beyond gender or ethnicity - whether age, economic, racial, or educational. The result? Missed potential and outdated hiring practices that slow companies down.

TrueView reimagines the interview process through intelligent analysis. Using AI-powered video, audio, and language insights, it identifies patterns — like interruptions, tone shifts, or linguistic biases — that impact hiring objectivity. The platform translates these signals into actionable feedback, enabling organizations to make data-driven, fair, and future-focused decisions.

By bringing transparency and behavioral intelligence into hiring, TrueView isn’t just a tool — it’s a catalyst for smarter, tech-driven talent evaluation that helps companies evolve faster and hire better.

What it does

TrueView is an AI-powered interview platform designed to detect and reduce bias in real time. It records and analyzes interviews using video, audio, and language data, identifying subtle indicators of bias such as:

  • Unequal speaking time or interruptions
  • Tone disparities between interviewers and candidates
  • Emotionally charged or gendered language

After each interview, TrueView generates a comprehensive bias and performance report that highlights potential red flags and provides actionable insights for interviewers.

Displays customized insights to each stakeholder:

  • HR sees both performance and bias analysis.
  • Interviewers see their feedback evaluated for tone and fairness.
  • Candidates receive a concise summary of their strengths and growth areas.

Tech Stack Overview

  • Frontend: React + TypeScript + Tailwind + ShadCN UI
  • Backend: Flask (Python) REST API, meetings server hosted on Render
  • AI Models:
    • Whisper → Audio transcription
    • Gemini 2.5 Flash → Bias, tone, and fairness analysis
  • Video Processing: MediaRecorder API + FFmpeg
  • Storage: Local JSON + mock HR data (extendable to Firebase/PostgreSQL)

Workflow

  • Candidate and interviewer join the same interview room via WebRTC.
  • The interview is recorded locally using the MediaRecorder API.
  • After the interview ends, the video is uploaded to the Flask backend.
  • Whisper transcribes the audio.
  • The transcript is analyzed by Gemini, which returns structured bias and tone metrics.
  • The frontend displays personalized AI reports for HR, interviewer, and candidate.

Challenges we ran into

  • Audio Conversion Issues: FFmpeg occasionally failed when processing large .webm files, requiring multiple encoding retries.
  • Prompt Engineering for Gemini: Getting balanced, JSON-structured AI responses demanded iterative tuning of prompts for fairness detection.
  • Role-Based Access: Ensuring that HR, interviewers, and candidates only saw their respective data required careful local state and localStorage handling.
  • Bias Quantification: Translating subjective fairness concepts (tone, interruptions) into numerical bias scores was a major design challenge.
  • Frontend Integration: Seamlessly merging AI results into ShadCN components while keeping the UX simple and accessible.
  • Deployment on Render: Integrating the frontend with the server backend on Render introduced CORS and routing complications. Configured proxy routes, API endpoints, and managed environment variables to ensure smooth communication between services.
  • Debugging build failures and aligning Node.js and Python environments for production took extensive trial and error but resulted in a stable multi-service deployment pipeline.

Accomplishments we’re proud of

  • Built a complete AI-driven bias detection system in under 10 hours.
  • Achieved real-time transcription + AI fairness scoring through Gemini’s API.
  • Designed an HR dashboard that visualizes performance and bias trends.
  • Created distinct views for candidates, interviewers, and HR for maximum transparency.
  • Delivered an intuitive interface with live camera feed, chat, and note-taking tools.

What’s next

  • Scalable Cloud Backend: Migrate from local Flask to Firebase or AWS for enterprise-grade deployment.
  • Zoom / Teams Integration: Automatically analyze live interviews through plugins.
  • Expanded AI Metrics: Detect microaggressions, emotional imbalance, and diversity sentiment.
  • Privacy Mode: End-to-end encrypted anonymization for all recordings.
  • Fairness Analytics Dashboard: Track diversity, bias trends, and progress over time.

Built With

Share this project:

Updates