How to Configure the Hiring Agent for Development

Set DEVELOPMENT_MODE = True in config.py, configure your .env file with LLM credentials, install dependencies from requirements.txt, and run python score.py <pdf> to generate cached JSON payloads and a resume_evaluations.csv log.

The Hiring Agent from interviewstreet/hiring-agent supports a development configuration that preserves intermediate results and caches expensive API calls. This mode accelerates iteration when tuning prompts or switching LLM providers by writing evaluation logs to CSV and storing conversion results in local JSON caches.

Enable the Development Flag

In config.py, set the boolean flag that controls caching and export behavior:

# config.py

DEVELOPMENT_MODE = True  # enables caching and CSV export

When active, this flag instructs score.py (lines 95–96 in the source) to write the evaluation CSV and cache JSON results after each run. Disabling it (False) removes these artifacts and forces fresh processing on every execution.

Configure Runtime Environment

Copy the example environment file to define your LLM backend and model parameters:

cp .env.example .env

Edit .env to select between Ollama or Gemini providers:

LLM_PROVIDER=ollama
DEFAULT_MODEL=gemma3:4b

# GEMINI_API_KEY=your_key_here  # Required only for Gemini

The models.py module instantiates the appropriate client based on these variables, allowing you to swap backends without modifying the pipeline code.

Install Dependencies

Create a virtual environment and install the required packages listed in requirements.txt:

python -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate

pip install -r requirements.txt

Execute the Pipeline

Run the scorer against a PDF resume:

python score.py path/to/resume.pdf

When DEVELOPMENT_MODE is True, the script automatically writes resume_evaluations.csv to the project root and populates the cache/ directory with intermediate JSON files.

Verify Development Outputs

Inspect the generated artifacts to confirm caching is active and reusable:

• CSV Log: resume_evaluations.csv contains a timestamped record of every evaluation.
• JSON Caches: Files like cache/resumecache_<basename>.json and cache/githubcache_<basename>.json store PDF-to-Markdown conversions and GitHub API responses.

These caches prevent redundant processing and API calls during iterative development, allowing you to inspect or reuse data without re-hitting external services.

Summary

• Set DEVELOPMENT_MODE = True in config.py to toggle caching and CSV export.
• Configure .env with LLM_PROVIDER and model details to select your backend without code changes.
• Install packages via requirements.txt before running the pipeline.
• Execute python score.py <pdf> to generate caches and evaluation logs automatically.
• Review the cache/ directory and resume_evaluations.csv to verify intermediate results are persisted.

Frequently Asked Questions

What is the purpose of DEVELOPMENT_MODE in the Hiring Agent?

When set to True, the flag instructs score.py to write a resume_evaluations.csv log and cache JSON payloads in the cache/ directory. This eliminates redundant PDF processing and GitHub API calls, significantly speeding up debugging cycles when iterating on prompts or extraction logic.

How do I switch between Ollama and Gemini LLM providers?

Edit the .env file (copied from .env.example) and set LLM_PROVIDER to either ollama or gemini. For Gemini, you must also provide GEMINI_API_KEY. The models.py module reads these variables at runtime to instantiate the correct client without requiring changes to the pipeline code.

Where are the cached files stored during development?

Cached JSON files are written to the cache/ directory relative to the project root. You will find resumecache_<basename>.json for PDF conversions and githubcache_<basename>.json for GitHub enrichment data. The CSV log resume_evaluations.csv is written to the root directory.

Can I disable caching once the development configuration is set?

Yes. Change DEVELOPMENT_MODE = False in config.py before running score.py. When disabled, the pipeline will not write CSV logs or JSON cache files, and it will re-process PDFs and query the GitHub API on every execution.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}