AI Resume Matcher

This is a Flask application that matches resumes (PDF/Text) to job descriptions using Mistral AI.

Setup

1. Install Dependencies:

   cd app
   python3 -m venv venv
   source venv/bin/activate
   pip install -r requirements.txt

2. Configure API Key: Edit app/.env and add your Mistral API Key:

   MISTRAL_API_KEY=your_actual_key_here
   

3. Data: Ensure job_descriptions.csv is in the project root (../job_descriptions.csv relative to app/).

Running

cd app
python app.py

Run the app with custom CSV / Embeddings paths

You can override the CSV and embeddings binary used by the app at runtime:

# Example: use a trimmed CSV and small embeddings
.venv/bin/python src/app.py --csv data/job_descriptions_small.csv --embeddings data/job_embeddings_small.bin --port 5001

The app will print startup messages showing which CSV and embeddings were loaded.

Extract a smaller CSV from the large zip

If you have data/job-description-dataset.zip (the original large archive), you can extract just the first 50 rows into data/job_descriptions_small.csv for faster local testing:

cd project_root
.venv/bin/python src/extract_small_csv.py

Then in src/app.py, change the CSV path to data/job_descriptions_small.csv if you want to run with the trimmed dataset. Or you can replace the big file with the small one (be careful to backup first).

Usage

1. Open http://localhost:5000 in your browser.
2. Upload a resume (PDF or Text).
3. The app will:
   • OCR the resume (if PDF) using Mistral OCR.
   • Extract structured data (JSON) using Mistral Large.
   • Generate embeddings for the resume.
   • Match against the top 20 jobs from the CSV (limit set for demo).
   • Return the parsed resume and top matches.

Testing & troubleshooting

See docs/TESTING.md for a step-by-step guide (commands and example files) that reproduces the E2E testing we used to validate API endpoints, embeddings, and the matching logic. The doc includes fallback options if you hit Mistral rate limits and instructions for saving unique test artifacts under data/.

Architecture

• Parsing: Mistral OCR + Mistral Large (Chat Completion) for JSON extraction.
• Matching: Mistral Embeddings + Cosine Similarity.

API Endpoints

• POST /match
  • Upload a resume file (PDF or text) as file. Returns resume_analysis and top_matches where top_matches is a list of {index, score} referencing rows in data/job_descriptions.csv.
• GET /jobs/
  • Returns a single job by its index (zero-based by CSV order).

External resources

These are the datasets and Mistral API docs referenced in the original project prompt and used while building/testing the system.

• Job listings dataset (Kaggle): https://www.kaggle.com/datasets/ravindrasinghrana/job-description-dataset/data
• Resumes dataset (Hugging Face): https://huggingface.co/datasets/datasetmaster/resumes
• Mistral Document Annotations API (Document AI): https://docs.mistral.ai/capabilities/document_ai/annotations
• Mistral Text Embeddings API: https://docs.mistral.ai/capabilities/embeddings/text_embeddings

Please consult these links for dataset schema details and Mistral API usage when generating embeddings and parsing documents.