An AI-powered defence recruitment platform that conducts autonomous interview sessions with real-time face verification, NLP-based answer evaluation, tone-adaptive questioning, and OS-level kiosk lockdown. Simulates a full SSB interview panel.
4-Signal NLP Scoring Engine
Evaluates answers using a weighted composite of Keyword Match (35%), Semantic Similarity via TF-IDF cosine (25%), Completeness (20%), and Coherence (20%), producing a 0–10 score per response.
Tone-Adaptive Questioning
Detects candidate behavioral tone (confident, nervous, evasive, aggressive, thoughtful) in real-time and dynamically adjusts question difficulty. Evasive candidates get direct probes, nervous candidates get familiar topics.
Face Verification & Anti-Spoofing
ArcFace identity verification with multi-layered liveness: blink detection (EAR), head pose estimation (solvePnP), motion analysis to counter photo attacks, and multi-face detection for unauthorized person flagging.
4-Layer Authentication
RFID + Roll Number + Date of Birth + Live Face Verification, simulating government-grade NIC database validation with permanent RFID-to-account binding.
OS-Level Kiosk Lockdown
C# binary with Win32 low-level keyboard hooks blocks Alt+Tab, Win key, Ctrl+Esc. PowerShell-driven taskbar hiding + Chrome kiosk-mode turns a standard Windows machine into a secure exam terminal.
Architecture
React Frontend
Candidate UI · Dashboard · Interview
Node.js Backend
Auth · Sessions · Orchestration
NLP Engine
HuggingFace · TF-IDF · Tone Detection
Face AI API
ArcFace · MTCNN · Liveness
Kiosk Security
C# · PowerShell · Win32 Hooks
MongoDB
Users · Sessions · Reports
Data Flow
Designed the 4-service microservice architecture: React frontend, Node.js orchestrator, Python NLP engine, and Python Face AI API
Built the NLP scoring engine with weighted 4-signal evaluation and tone-adaptive question selection using HuggingFace Transformers
Implemented the face verification pipeline with ArcFace, multi-layered liveness detection, and anti-spoofing measures
Developed the 4-layer authentication system simulating government-grade NIC validation
Created the OS-level kiosk lockdown using C# Win32 API hooks and PowerShell automation