Veritas - Interview Practice Tool

Practice ANYTHING with AI to gain confidence, prepare for:interviews, speechs, or crucial conversations. Utilizes ML + LLMs to analyze body language, vocal patterns and content of interview.

Comment

InterviewPro: AI-Powered Interview Intelligence

Inspiration

InterviewPro is an AI interview simulator that helps candidates practice realistic, role-specific interviews and get actionable feedback in minutes. The idea came from recognizing that interview preparation is often expensive, time-consuming, or purely theoretical. We wanted to build something that enables people to become better communicators through realistic practice and multimodal feedback. Although right now the app is geared towards interviews, we see a plethora of potential use cases. Two examples of potential use cases are:

B2B product for retail companies to help customer facing employees practice common interactions, speaking about the product to increase sales. Help people who are learning new languages to better understand cultural communication signals.

What It Does

Users enter their target role, company, and experience level, then run a live mock interview with an adaptive interviewer. After each session, the app generates a structured performance report with communication clarity, technical depth, behavioral signal strength, and prioritized improvement suggestions. Instead of only scoring transcript quality, InterviewPro analyzes delivery signals like voice prosody, pacing, pauses, filler-word patterns, facial expression, and body language. It is our hope that this technology will enable people to become better communicators. We are beginning to build out support for packs which will enable users to further customize their experience. An example of a pack may be the JP Morgan interview process where users can practice each interview stage in accordance with the publicly released interview order/structure released by JPM.

How We Built It

It is implemented with a React frontend, TypeScript/Node backend, and managed data/storage services for session state and recordings. We also integrate third-party services for voice/video delivery and workflow orchestration; those integrations are infrastructure support, while Gemini 3 remains central to interview generation and analysis. We use Gemini 3 for TTS during the interview, through the Eleven Labs Agents SDK. Gemini 3 is the reasoning engine of the app driving post-interview analysis. In some cases Groq inference of oss 120b is used for fast generative UI which enables users to customize their experience.

Challenges We Faced

I am a nontechnical founder, vibecoder, so making this app was no easy task. I used a combination of Antigravity, Gemini CLI, Claude Code and Codex to get the app done, while using n8n for some backend functionality that I wanted finegrained control over. I used tools like remotion to create my product video. When I run into an error sometimes it is very difficult to fix because I am unable to perfectly diagnose the cause of the error which often leads to frustration.

Integrating ML systems for voice & facial analysis proved to be very difficult as well, I was very commited to only intergrating features that had a very high accuracy threshold so testing different ML systems was very tedious.

What I Learned

My multi-agent orchestration abilties improved a ton working on this project, learning when and when not to use swarms enabled me to speed through some parts of the project and slow down and focus the models when working on critical details.

While doing market research and getting product feedback, I realized a lot of my expectiations for how users would use the product were not aligned with user needs and I had to pivot a lot of features.

Built With

Share this project:

Updates