Skip to content

About

Bias-Resistant Resume Ranker uses Python, FastAPI, and advanced NLP (GPT-based) models to score and rank resumes while detecting and mitigating bias. It features modular microservices, custom fairness algorithms, and asynchronous processing, ensuring scalable, explainable, and fair candidate evaluation through a user-friendly interface.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

2 Commits

Folders and files

Repository files navigation

Bias-Resistant Resume Ranker

A full-stack, AI-powered talent intelligence platform designed to help recruiters screen candidates fairly and efficiently, while also empowering candidates to improve their resumes.


🎥 Live Demo

(It is highly recommended to add a GIF or short video here showing the application in action. You can use a tool like Giphy Capture or Kap to record your screen.)

Application Screenshot Placeholder


✨ Features

This application provides a comprehensive suite of tools for both recruiters and candidates:

For Recruiters:

  • Intelligent Resume Evaluation:
    • Skill-Fit Scoring: Uses a Sentence Transformer model to calculate a semantic similarity score between the resume and the job description.
    • LLM-Powered Skill Extraction: Leverages a local Llama 3 model via Ollama to intelligently identify technical and soft skills, going beyond simple keyword matching.
    • Resume Quality Analysis: The LLM provides scores (1-10) and justifications for resume quality based on:
      • Quantification: Use of metrics and numbers to demonstrate impact.
      • Clarity: Conciseness and readability.
      • Impact Language: Use of strong action verbs.
  • DEI & Fairness Auditing:
    • Bias Flagging: Automatically scans resumes for potentially biased language (gendered pronouns, age-related terms, etc.).
    • Batch DEI Audit: Upload a CSV of resumes to perform a batch fairness audit. The system calculates and visualizes key fairlearn metrics like Demographic Parity Difference and Equalized Odds Difference.
    • Historical Benchmarking: Single resume scores are compared against historical averages from batch audits to provide context on fairness.
  • Asynchronous Processing: The batch audit feature uses Celery and Redis to run as a background task, ensuring the UI remains responsive and can handle large datasets without timing out.

For Candidates:

  • Automated Bias Correction: Provides a rewritten, bias-neutral version of a resume to ensure it is judged purely on merit.
  • Side-by-Side Diff Viewer: Clearly highlights the changes between the original and rewritten resume text.
  • PDF Parsing: Supports direct upload of PDF resumes, with text automatically extracted for analysis.

🛠️ Tech Stack & Architecture

The application is a containerized, multi-service system orchestrated with Docker Compose.

graph TD
    subgraph "User's Machine"
        A[User's Browser] -->|HTTP Requests| B(Streamlit Frontend);
    end

    subgraph "Docker Environment"
        B -->|API Calls| C{FastAPI Backend};
        C -->|Cache & Tasks| D[Redis];
        C -->|Async Tasks| E(Celery Worker);
    end

    subgraph "Host Machine Services"
        E -->|LLM Calls| F(Ollama - Llama 3);
        C -->|LLM Calls| F;
    end

    style B fill:#268bd2,stroke:#333,stroke-width:2px
    style C fill:#f5a623,stroke:#333,stroke-width:2px
    style D fill:#d33682,stroke:#333,stroke-width:2px
    style E fill:#f5a623,stroke:#333,stroke-width:2px,stroke-dasharray: 5 5
    style F fill:#859900,stroke:#333,stroke-width:2px
Loading
  • Frontend: Streamlit - For a fast, interactive user interface.
  • Backend: FastAPI - A high-performance Python framework for building the API.
  • AI / LLM: Ollama & Llama 3 - Runs a powerful open-source LLM locally for all generative AI tasks.
  • Embeddings: Sentence-Transformers - For generating semantic embeddings for skill-fit scoring.
  • Fairness: Fairlearn - Used in the DEI Audit to calculate fairness metrics.
  • Task Queue: Celery - For handling long-running batch processing jobs asynchronously.
  • Caching & Messaging: Redis - Serves as both a cache for embeddings and a message broker for Celery.
  • Containerization: Docker & Docker Compose - To ensure the application is portable, scalable, and easy to run.

🚀 Getting Started

To run this application, you will need Docker Desktop and Ollama installed on your machine.

  1. Install Ollama:

  2. Pull the Llama 3 Model:

    • Run the following command in your terminal:
      ollama run llama3
  3. Clone the Repository:

    git clone https://github.com/your-username/Bias-Resistant-Resume-Ranker.git
    cd Bias-Resistant-Resume-Ranker
  4. Launch the Application:

    • Make sure Docker Desktop and Ollama are running.
    • Run the following command from the root of the project directory:
      docker-compose up --build
    • This command will build the Docker images, start all the services (frontend, backend, Redis, and Celery worker), and connect them.
  5. Access the Application:


📝 API Endpoints

The FastAPI backend exposes the following key endpoints:

  • POST /score_resume: Scores a single resume against a job description.
  • POST /suggest_corrections: Rewrites a resume to be bias-neutral.
  • POST /batch_audit: Submits a batch job to audit a CSV of resumes for fairness.
  • GET /audit_status/{task_id}: Checks the status of a background audit task.

This project was built with the assistance of an AI pair programmer.

curl http://host.docker.internal:11434

About

Bias-Resistant Resume Ranker uses Python, FastAPI, and advanced NLP (GPT-based) models to score and rank resumes while detecting and mitigating bias. It features modular microservices, custom fairness algorithms, and asynchronous processing, ensuring scalable, explainable, and fair candidate evaluation through a user-friendly interface.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages