Adversarial Multi-Agent Framework for AI-Driven Resume Optimization

This project presents a sophisticated AI system composed of multiple intelligent agents designed to generate highly tailored and optimized resumes. By leveraging a multi-agent architecture, advanced prompt engineering techniques, and an adversarial feedback loop, this framework helps users create resumes that are not only ATS-compliant but also perfectly aligned with specific job descriptions and company cultures.

Overview

In today's competitive job market, a generic resume is not enough. This project tackles the challenge of resume personalization by creating an intelligent pipeline that analyzes a candidate's profile (CV, GitHub), a target job description, and company details to produce a superior, customized resume.

The system employs a series of specialized AI agents, each with a distinct role, from parsing job requirements to analyzing a candidate's portfolio and generating impactful, STAR-formatted bullet points. The core of the system is an adversarial loop between a Resume Builder Agent and an ATS Checker Agent, which iteratively refines the resume to achieve a high ATS score and strong keyword alignment.

Key Features

• Multi-Agent Architecture: A team of specialized AI agents work together to handle different aspects of resume creation, from data extraction to content generation and evaluation.
• Adversarial Feedback Loop: An innovative generative-evaluative loop between the Resume Builder and ATS Checker agents ensures continuous improvement and optimization of the resume.
• Advanced Prompt Engineering: Utilizes a variety of prompt engineering strategies, including Chain of Thought (CoT), Step-Back Reasoning, and Chain of Verification (CoV), to enhance the reasoning and output quality of the LLMs.
• Deep Profile Analysis: Goes beyond simple keyword matching by analyzing GitHub repositories, company culture, and the candidate's existing CV to create a holistic profile.
• ATS Compliance and Optimization: Generates resumes that are rich in relevant keywords and formatted to pass through modern Applicant Tracking Systems.
• Interactive Streamlit UI: A user-friendly web interface for easy input of information and visualization of the final generated resume.
• Secure and Organized: The project is structured for clarity and maintainability, with all sensitive information like API keys managed through a .env file.

System Architecture

The system is designed as a modular, multi-agent pipeline that takes four key inputs from the user:

1. Job Description: The description of the job the user is applying for.
2. Company Name: The name of the company.
3. GitHub Profile: The user's GitHub username or profile URL.
4. Existing CV: The user's current CV in PDF or TXT format.

The pipeline consists of the following agents:

• Agent 1: Job Description Parser: Extracts key responsibilities, skills, and requirements from the job description.
• Agent 2: Employer Data Extractor: Gathers information about the company's culture, values, and tech stack from public sources.
• Agent 3: GitHub Profile Analyzer: Analyzes the user's GitHub profile to identify key projects, technologies used, and contributions.
• Agent 4: Portfolio Analyzer: Synthesizes the information from the CV and GitHub to create a comprehensive profile of the candidate.
• Agent 5: Resume Builder: Generates the resume content, including impactful, STAR-formatted bullet points, based on the analyzed data.
• Agent 6: ATS Checker: Evaluates the generated resume for ATS compliance, keyword density, and overall quality, providing feedback for refinement.
• Conditional Input Agent: Checks for missing or insufficient input data and prompts the user for more information if needed.

Technical Stack

• Backend: Python
• AI Framework: CrewAI
• LLMs: OpenAI GPT-4.1 (or other models like Groq Llama)
• Frontend: Streamlit
• PDF Processing: PyPDF2, ReportLab
• Web Scraping: BeautifulSoup
• Dependencies: See requirements.txt for a full list of dependencies.

Prompt Engineering Strategies

To maximize the effectiveness of the LLMs, this project employs several advanced prompt engineering techniques:

• Chain of Thought (CoT): Used to break down complex tasks like job description analysis into a series of intermediate reasoning steps.
• Step-Back Reasoning: Allows agents to self-critique and refine their generated content.
• Self-Consistency: Ensures that the output from the agents is logically coherent and consistent.
• Chain of Verification (CoV): Used to validate information scraped from public sources about the company.

Project Structure

The project is organized into the following directory structure:

• src/: Contains the main source code for the application.
  • main.py: The main script to orchestrate the entire resume generation workflow (can be run as a standalone script).
  • streamlit_app.py: The Streamlit web interface.
  • agents/: Contains the scripts defining the AI agents (final4.py, next_agent.py).
  • utils/: Contains utility scripts for tasks like PDF conversion, web scraping, and input validation.
• data/: Stores the input data files provided by the user.
• output/: Stores the generated output files, including the final resume PDF and intermediate text files.
• knowledge/: Contains the knowledge base used by the AI agents.
• requirements.txt: A list of all the Python dependencies for this project.
• .env: For storing API keys and other sensitive information.

Setup and Installation

1. Clone the Repository:

   git clone <repository-url>
   cd <repository-directory>

2. Set up the Environment:
   • Create a virtual environment:

     python -m venv venv

   • Activate the virtual environment:

     venv\Scripts\activate

   • Install the required dependencies:

     pip install -r requirements.txt

3. Provide API Keys:
   • Create a .env file in the root of the project.
   • Add your OpenAI API key to the .env file:

     OPENAI_API_KEY="YOUR_API_KEY_HERE"
     

How to Run

• Start the Streamlit app:

  streamlit run src/streamlit_app.py

• Fill in the required information in the web interface and click "Process Application".
• Once the initial processing is complete, a "Generate Resume" button will appear. Click it to run the full crewai pipeline.
• The final, tailored resume will be displayed in the app and saved as output/resume.pdf.

Future Work

• Real-time Job Market Data Integration: Integrate with APIs from platforms like LinkedIn or Glassdoor to provide real-time suggestions based on market trends.
• Multilingual Support: Extend the system to support languages other than English.
• Advanced Personalization: Implement adaptive learning agents that tailor their feedback based on the user's career level.
• Collaborative Module: Add a feature for mentors, peers, or recruiters to comment on and co-edit resumes in real-time.

Authors

• P Kavya Samhitha
• Piyush Sahoo
• Pratyush Tripathi

Acknowledgments

We would like to express our sincere gratitude to our faculty guide, Dr. Pooja Agarwal, for her invaluable guidance, insights, and support throughout the course of this project.