Job Application Research & Tailoring Agent

An AI-powered multi-agent pipeline built with LangGraph, LangChain, and OpenRouter to research companies and tailor CVs for specific job descriptions with built-in hallucination protection.

Features

• JD Ingestion & Validation: Supports URLs and local files. Automatically verifies if the content is a valid job description.
• Parallel Intelligence: Simultaneously researches the Hiring Company, Role Best Practices, and LinkedIn Competing Candidate Profiles (X-Ray Sourcing).
• Self-Refining Loop: Evaluates research for gaps and recursively refines data (up to 3 iterations).
• CV Sanitization: A dedicated auditor node that compares the tailored CV against your Source of Truth (YAML) to detect and remove hallucinations.
• Multi-Tiered Models: Uses faster models (GPT-4o mini) for research and stronger models (GPT-5 mini) for critical synthesis and auditing.
• Organized Outputs: Chronological, timestamped directories for every run.

Workflow Architecture

The agent uses a parallel fan-out/fan-in pattern with a self-correcting refinement and sanitization loop:

Setup

1. Install Dependencies:

   pip install -r requirements.txt

2. Configure Environment: Create a .env file in the root directory:

   OPENROUTER_API_KEY=your_openrouter_api_key
   TAVILY_API_KEY=your_tavily_api_key
   MODEL_VERSION=openai/gpt-4o-mini
   STRONG_MODEL_VERSION=openai/gpt-5-mini

Usage

Tailor a CV

Uses your master YAML CV and targets a specific job:

python main.py --cv data/base_cv/master_cv.yaml --jd "https://company.com/job-url"

Evaluation Suite

Run a parallel, semantic hallucination check against all JDs in data/eval_jd/:

make evaluate

Utility Commands

Command	Description
make lint	Run Ruff linting and formatting checks
make test	Run the unit test suite
python main.py --visualize	Generate the workflow.png diagram
python main.py --pdf <file.md>	Render an existing Markdown file to a styled PDF

CLI Arguments

Argument	Description	Default
--cv	Path to your original CV (YAML, PDF, or TXT)	data/base_cv/master_cv.yaml
--jd	URL or path to the Job Description	data/eval_jd/apple_mle.yaml
--model	Primary model for research	openai/gpt-4o-mini
--personalize	Custom instructions for the LLM	None

Outputs

Each run saves to data/output/YYYYMMDD_HHMMSS/:

• TAILORED_CV.pdf: The final professionally formatted resume.
• STRATEGY_REPORT.pdf: Detailed research and tailoring strategy.
• metadata.json: Logs the inputs, model versions, and audit status.