Fairscan

An Ontology-Driven Resume Screening Framework for Bias-Free, Context-Autonomous Recruitment

🚀 Overview

Fairscan revolutionizes AI-powered resume screening by eliminating external biases and ensuring complete organizational context autonomy. Unlike traditional systems that rely on generic pre-trained models, Fairscan leverages semantic reasoning and organization-specific ontologies to provide transparent, interpretable, and contextually relevant candidate evaluations.

✨ Key Features

1. 🎯 Bias-Free Screening: Eliminates external biases from pre-trained models

2. 🏢 Organizational Autonomy: Operates entirely within your infrastructure

3. 🧠 Cache Augmented Generation: Novel CAG methodology for efficient semantic processing

4. 📊 SBERT Integration: Advanced semantic similarity matching

5. ⚡ KV Cache Optimization: Fast ontology-driven context retrieval

6. 🔒 Privacy-First: Complete data sovereignty with no external dependencies

7. 📊 Transparent Decisions: Interpretable screening results with clear reasoning

8. ⚙️ Customizable Ontologies: HR-defined contexts for job requirements and cultural values

🎯 Problem Solved

Traditional AI resume screening systems face three critical challenges:

1. Bias Propagation: Inherit irrelevant biases from external training datasets
2. Privacy Vulnerabilities: Risk exposure of sensitive candidate information
3. Context Misalignment: Fail to capture organization-specific requirements and values Fairscan addresses these issues through semantic knowledge representation that accurately captures and utilizes organization-specific hiring contexts.

🏗️ Architecture

Fairscan implements a novel Cache Augmented Generation (CAG) methodology that combines semantic reasoning with efficient knowledge retrieval:

Core Components:

1.📥 HR Input Layer

PDF/DOCX resume processing Ontology document management

2.🔄 Document Processing Engine

Text extraction with privacy obfuscation Personal data anonymization

3.🧠 Ontology Processing

KV Cache generation for semantic efficiency Context window preloading Job description extraction and structuring

4.🎯 Job Description Matching

SBERT-based semantic similarity scoring Best job classification and ranking Multi-criteria evaluation framework

5.⚡ Cache Augmented Generation (CAG)

Resume text + matched job description fusion Ontology KV Cache integration for contextual reasoning Semantic analysis optimization

6.🤖 Open Source LLM Integration

Privacy-preserving local inference Semantic analysis and intelligent scoring Bias-free evaluation pipeline

7.📊 Results & Review System

Criteria-based scoring with transparency Detailed reasoning and recommendations HR review integration with audit trails

Project Requirements

System Requirements

• GPU Memory: 6GB VRAM minimum
• CUDA: Enabled environment required
• MySql Workbench 8.0

Software Requirements

• Python: 3.10
• llama-cpp-python: Latest version
• PyTorch: Latest with CUDA 12.4 support
• Torchvision: Latest with CUDA 12.4 support

Installation Steps

1. Set Environment Variables (Windows PowerShell)

$Env:LLAMA_CUBLAS = "1"
$Env:FORCE_CMAKE = "1"

# Add path_to_your NVIDIA GPU Computing Toolkit

# for example

$Env:CMAKE_ARGS="-DGGML_CUDA=on -DCMAKE_GENERATOR_TOOLSET=cuda='C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1'"

2. Install llama-cpp-python

pip install llama-cpp-python --upgrade --force-reinstall --no-cache-dir

3. Install PyTorch with CUDA Support

pip install torch torchvision --index-url https://download.pytorch.org/whl/cu124

4. Install the requirements.txt

pip install -r ./requirements.txt

5. Redis Server Configuration

Windows (WSL - Windows Subsystem for Linux)

1. Install WSL if not already installed

# Run in PowerShell as Administrator
wsl --install

2. Install Redis in WSL

# Update package list
sudo apt update

# Install Redis
sudo apt install redis-server

# Start Redis service
sudo service redis-server start

# Enable Redis to start on boot
sudo systemctl enable redis-server

3. Test Redis Installation

# Test Redis connection
redis-cli ping
# Should return: PONG

Downoad Models

Download all-MiniLM-L6-v2

# Create the directory if it doesn't exist
mkdir -p scan/main_engine/models

Download all-MiniLM-L6-v2 model to the specified directory scan/main_engine/models

huggingface-cli download sentence-transformers/all-MiniLM-L6-v2 --local-dir scan/main_engine/models/all-MiniLM-L6-v2

Prerequisites

Make sure you have the required packages installed:

pip install huggingface-hub sentence-transformers

Download Mistral-7B-Instruct-v0.3-Q3_K_L.gguf model to the specified directory scan/main_engine/models

https://huggingface.co/lmstudio-community/Mistral-7B-Instruct-v0.3-GGUF/resolve/main/Mistral-7B-Instruct-v0.3-Q3_K_L.gguf

How to Run the System

Setup and Execution Steps

1. Run database migrations

   python manage.py makemigrations
   python manage.py migrate

2. Create superuser account

  python manage.py createsuperuser

3. Start Celery workers

  celery -A fairscan worker --loglevel=info -E -Q chainprocessing --pool=threads

4. Start the development server

  python manage.py runserver

5. Access the application Open your browser and navigate to http://localhost:8000

6. Login to the system Use your superuser credentials created in step 2

7. Upload ontology document You can download a template using the "Download Template" button Upload your ontology document

8. Upload resume dataset Upload resumes (I used the dataset from Kaggle [https://www.kaggle.com/datasets/palaksood97/resume-dataset])

9. Process the data Click the "Process" button to begin analysis