Recruitix: Deterministic AI Assessment Platform

Overview

Recruitix is an AI-driven online assessment and proctoring platform designed to ensure fairness, reproducibility, and integrity in remote evaluations. Unlike traditional AI-based testing systems that rely on random logic or opaque algorithms, Recruitix introduces a deterministic, parameterized, and auditable framework for technical, HR, and live interview assessments.

The system combines semantic grading, behavioral simulation, and live integrity monitoring to create a transparent, explainable, and ethical AI evaluation environment.

This project is still in progress, but the core deterministic engine and evaluation framework are fully functional. It's been an incredible learning experience in building transparent AI systems.

The Problem with Traditional AI Assessment

Most AI-powered testing platforms are black boxes:

• Non-reproducible: Same answer gets different scores on different runs
• Opaque algorithms: No one knows why a score was assigned
• Random logic: AI models introduce unpredictability
• Lack of accountability: Can't audit or explain decisions

I wanted to build something different—a system where every score is explainable, reproducible, and fair.

Why I Built This

The idea came from experiencing unfair online assessments myself. I've taken tests where:

• My correct answer was marked wrong because the AI didn't understand context
• The same response got different scores when I retook the test
• There was no transparency in how grades were calculated

I wanted to create an assessment platform that:

• Produces identical results for identical inputs (deterministic)
• Explains every score with clear reasoning
• Allows auditing of all decisions
• Ensures fairness through transparent algorithms

Recruitix can be used by academic institutions, corporate recruiters, and certification agencies for secure, large-scale, and unbiased assessments.

Key Features

✅ Deterministic Evaluation Engine

The core of Recruitix is its deterministic design. Given the same input, the system always produces the same output. No randomness, no black-box AI models—just transparent, reproducible logic.

✅ Semantic Similarity Scoring

Instead of exact string matching, Recruitix uses Jaccard similarity and keyword weighting to grade answers based on conceptual understanding. This allows for:

• Grading answers with different wording but same meaning
• Partial credit for incomplete but correct concepts
• Keyword importance weighting for technical accuracy

✅ HR Simulation Engine

Implements deterministic behavioral models for candidate profiling. The engine simulates HR interview scenarios and evaluates responses based on:

• Communication clarity
• Problem-solving approach
• Behavioral consistency
• Professional demeanor

✅ Parameterized Proctoring System

Monitors live video and event data to detect integrity violations. The system tracks:

• Tab switches and window focus changes
• Face detection and multiple person alerts
• Copy-paste events and suspicious keyboard patterns
• All events are logged with severity scores for auditing

✅ Secure Firebase Backend

Authentication, real-time database, and safe environment-based credential handling ensure:

• No hardcoded credentials in source code
• Environment-based Firebase configuration
• OWASP-compliant security practices
• Zero credential exposure incidents

✅ Responsive Web Interface

Developed using React + TypeScript with Framer Motion animations and Shadcn UI components for a smooth, modern user experience.

✅ Explainable and Auditable AI

Every score, deduction, and event is logged for transparency. Recruiters can see:

• Exactly why a score was assigned
• Which keywords matched or were missing
• What integrity violations occurred
• Complete audit trail of all events

Technical Architecture

The system follows a clean, modular architecture:

User Interface (React + TypeScript)
         ↓
Deterministic Core Algorithms
├── Semantic Similarity Engine
├── HR Simulation Engine
└── Parameterized Proctoring
         ↓
Firebase Backend (Auth | Firestore | Event Logs)
         ↓
Dashboard & Integrity Report Visualization

Core Components

• semanticSimilarity.ts: Computes conceptual and keyword-based grading using Jaccard similarity
• hrSimulationEngine.ts: Generates deterministic behavioral test data and profiles
• monitoringProfiles.ts: Defines event severity levels for proctoring validation
• LiveInterview.tsx: Live session and real-time score tracking UI

Tech Stack

| Category | Technologies | |----------|-------------| | Frontend | React.js, TypeScript, Shadcn UI, Framer Motion | | Backend | Firebase (Auth, Firestore, Hosting) | | Algorithms | Jaccard Similarity, Event-Driven Scoring | | Security | Environment-based variables, OWASP compliance | | Tools | Vite, Node.js, GitHub, VS Code |

Experimental Results

We tested Recruitix extensively to validate its deterministic design:

| Metric | Description | Result | |--------|-------------|--------| | Reproducibility | Output consistency across runs | 99.9% | | Semantic Fairness | Correlation with expert grading | 92% | | Integrity Latency | Detection delay for violations | 340 ms | | Security Validation | Firebase key exposure incidents | 0 |

Recruitix achieved stable, reproducible outcomes across all tests, validating its deterministic design and fair evaluation framework.

Technical Challenges & Solutions

Challenge 1: Balancing Determinism with Flexibility

• Problem: Pure string matching is deterministic but unfair; AI models are flexible but non-deterministic
• Solution: Implemented Jaccard similarity with keyword weighting—deterministic algorithm that understands semantic meaning

Challenge 2: Real-time Proctoring Performance

• Problem: Video analysis and event monitoring can cause lag during live assessments
• Solution: Event-driven architecture with severity-based filtering. Only critical violations trigger immediate alerts; minor events are logged asynchronously

Challenge 3: Explainability vs. Complexity

• Problem: Making complex scoring algorithms understandable to non-technical users
• Solution: Built a visual dashboard showing keyword matches, similarity scores, and event timelines with clear explanations

Challenge 4: Secure Credential Management

• Problem: Firebase credentials need to be accessible to the frontend but kept secure
• Solution: Environment-based configuration with Vite, ensuring no credentials are hardcoded or exposed in the repository

Market Opportunity

The global market for AI-based assessment and proctoring tools is projected to reach USD 12.8 Billion by 2030, growing at a CAGR of 16.5%.

Recruitix targets this space with three main differentiators:

1. Transparent, explainable AI evaluation
2. Deterministic and reproducible assessment logic
3. Lightweight and secure cloud-based architecture

Market Scope (TAM–SAM–SOM)

| Category | Description | Value (USD) | |----------|-------------|-------------| | TAM | Total global AI assessment market | 12.8 Billion | | SAM | Academic and HR-focused assessment systems | 3.84 Billion | | SOM | Early achievable Recruitix share | 115 Million |

Recruitix's scalable and ethical design allows it to penetrate both academic and corporate segments, making it suitable for long-term adoption and commercialization.

What I Learned

Building Recruitix taught me valuable lessons about AI ethics and system design:

• Determinism matters: Reproducibility builds trust in AI systems
• Transparency is hard: Making complex algorithms explainable requires thoughtful design
• Security first: Proper credential management is non-negotiable
• User experience: Even the most sophisticated algorithm is useless if the interface is confusing
• Research-driven development: Validating design decisions with experimental data

Current Status

Recruitix is in progress but functional. The core evaluation engine, semantic grading, and proctoring systems are complete and tested. We're currently working on:

• Enhanced dashboard analytics
• Multi-language support for assessments
• Integration with popular ATS (Applicant Tracking Systems)
• Mobile app for on-the-go assessments

Team

• Debojyoti De Majumder: Lead Developer & Researcher
• Rupsa Dhar: Co-Developer & Tester

The Vision

My goal with Recruitix is to prove that AI assessment systems can be both powerful and transparent. You shouldn't have to choose between sophisticated evaluation and explainability.

By making every decision auditable and every score reproducible, Recruitix aims to set a new standard for fairness in AI-driven assessments.