Problem

The hiring-agent tool currently provides robust 1-to-1 resume analysis based on the implicit job description for HackerRank's SDE Intern role. However, two critical capabilities are missing:

1. The ability to evaluate candidates against an explicit job description provided as a PDF
2. Support for batch processing multiple candidates against a single job description, which reflects the typical N-to-1 shortlisting workflow used in real-world hiring

Adding these capabilities would significantly expand the utility of the existing agent and better align it with practical recruitment workflows.

Proposed Solution

I propose implementing two high-impact workflows to extend the CLI functionality:

1. JD-Augmented 1-to-1 Scoring

Enables users to provide both a resume and a job description for detailed fit analysis:

python score.py /path/to/resume.pdf /path/to/jd.pdf

2. Multi-Candidate Batch Processing and Shortlisting

Accepts a directory of resumes, a single job description, and a cutoff score threshold. Processes all resumes, ranks them, and outputs a shortlist of candidates meeting the minimum score:

python score.py /path/to/resume_directory /path/to/jd.pdf <cutoff_score_int>

Implementation Details

A comprehensive set of changes has been developed in a separate fork that integrates cleanly with the existing architecture:

Core Module Changes

models.py

• Added ScoresWithJD and EvaluationDataWithJD models to structure enriched evaluation (with JD) output

template_manager.py

• Introduced jd_system_message and jd_evaluation_criteria prompts for fit analysis

evaluator.py

• Implemented _load_evaluation_prompt_with_jd for loading the final, decorated JD-based evaluation prompt
• Added evaluate_resume_with_jd to handle the JD-based evaluation flow

transform.py

• Created transform_evaluation_response_with_jd to parse LLM output into CSV rows (to be saved)

score.py

• Significantly refactored the driver script to support new CLI argument patterns
• Added _jd_context function for job description parsing from PDF to text
• Implemented evaluate_resume_with_jd as the core 1-to-1 scoring function
• Created dir_main function to orchestrate batch processing and ranking workflow in the directory
• Added print_evaluation_results_with_jd for clean, user-facing output formatting like the base evaluation output
• Sanity checks have been added for all CLI usage arguments

Output (1-to-1):

Output (N-to-1) (snippet)

Status:
The complete implementation for this feature is finished on my fork. The code is modular, follows the project's style (black), and is ready for review. I am opening this issue to discuss the feature and ensure it aligns with the project's roadmap. If this direction is welcome, I am happy to add the required smoke tests and open a formal Pull Request immediately.