Buddy-ai-interview

Buddy AI — AI Interview Evaluation System

A complete AI-powered interview evaluation platform built with Flask, featuring semantic verbal scoring, real-time face proctoring, role-specific MCQs, and automated PDF performance reports.

Features

Module	Technology
Candidate Registration	Flask + SQLite, secure PDF resume upload
Technical MCQ Round	CSV-driven, role-specific, 6/10 progression gate
Verbal AI Interview	Web Speech API + SentenceTransformer (cosine similarity)
Face Proctoring	OpenCV Haar Cascade, webcam snapshot every 5s
Admin Dashboard	Secure portal, candidate search, FPDF2 PDF reports
Buddy Persona	Warm, empathetic AI guide between questions

Quick Start

# 1. Clone the repo
git clone https://github.com/<your-username>/buddy-ai-interview.git
cd buddy-ai-interview

# 2. Create a virtual environment
python -m venv venv
venv\Scripts\activate        # Windows
# source venv/bin/activate   # Mac/Linux

# 3. Install dependencies
pip install -r requirements.txt

# 4. Run the app
python app.py
# Open http://localhost:5000

Usage

Candidate

http://localhost:5000 → Start Interview
Fill registration form, upload PDF resume
Complete 10 MCQ questions (20-min timer) — need ≥ 6/10 to proceed
Complete 5 verbal questions via voice or typing — Buddy provides reflections
View results with MCQ score, verbal AI score, face alert log, and hire recommendation

Admin

http://localhost:5000/admin/login
Default credentials: admin / admin123
Search candidates, view full transcripts, download PDF reports

Project Structure

├── app.py                  Flask application factory
├── config.py               Configuration (roles, paths, auth)
├── models.py               SQLAlchemy models (Candidate, MCQResult, VerbalResult, FaceAlert)
├── questions.csv           45 MCQs (15 × 3 roles)
├── verbal_questions.json   15 verbal questions + ideal answers
├── routes/
│   ├── candidate.py        Registration, MCQ, Verbal, Results
│   ├── admin.py            Dashboard, candidate detail, PDF downloads
│   └── api.py              /api/verbal/score, /api/face/check
├── services/
│   ├── scorer.py           SentenceTransformer semantic scoring
│   ├── face_check.py       OpenCV face detection
│   └── pdf_report.py       FPDF2 report generation
├── templates/              9 Jinja2 HTML templates
└── static/
    ├── css/style.css       Dark premium design system
    └── js/                 MCQ controller, Verbal + Buddy flow, Face monitor

Tech Stack

Backend: Python 3.10+, Flask 3.x, Flask-SQLAlchemy, SQLite
AI/ML: sentence-transformers (paraphrase-MiniLM-L6-v2), scikit-learn
Vision: opencv-python (Haar Cascade face detection)
PDF: fpdf2
Frontend: Vanilla HTML/CSS/JS, Web Speech API

Notes

The SentenceTransformer model (~80 MB) downloads automatically on the first verbal scoring call. Web Speech API works best in Chrome/Edge with a working microphone. Candidates can also type answers directly if voice is unavailable.

License

MIT