AI-Powered Candidate and Vacancy Recommendation System
Scalable recommendation system using OpenAI models to match candidates with vacancies efficiently.
The AI-Powered Candidate and Vacancy Recommendation System is a cutting-edge solution designed to streamline recruitment processes. It uses OpenAI embedding models to analyze candidates’ resumes and job vacancies, offering precise recommendations. The similarity metric employed is cosine similarity, which ensures accurate semantic matching between candidates and vacancies. By leveraging embeddings, this solution scales efficiently to large datasets.
In comparative evaluations, the system achieved 88% alignment with ChatGPT’s direct recommendations, as measured by the Area Under the Curve (AUC). While ChatGPT provides highly accurate results, it is significantly more costly. The project includes a graphical representation of the AUC metric, demonstrating the strong performance of this scalable methodology against the more expensive ChatGPT approach.
Features
Candidate Matching
Leverage AI to recommend the most relevant candidates for a given vacancy, considering detailed profiles and job descriptions.
Vacancy Recommendations
Enable candidates to discover vacancies that align perfectly with their experience, skills, and preferences.
Technology Stack
Python
Primary language for embedding generation, data preprocessing, and API integration.
OpenAI Models
Embedding-based analysis of resumes and job descriptions to calculate semantic affinity.
NumPy & Pandas
Efficient data manipulation and preprocessing for large-scale candidate and vacancy datasets.
PostgreSQL
Robust database management for storing candidate and vacancy data securely.
Tesseract OCR
Extract text from uploaded resumes for embedding generation using Python and JavaScript integrations.
Project Achievements
The project culminated in the successful creation of a scalable recommendation system with the following highlights:
-
Optimal Model Configuration: The text-embedding-3-small model truncated to 256 dimensions provided the best balance of accuracy and cost.
-
High Accuracy: AUC score improved from 0.72 to 0.88 in the final testing phase, demonstrating strong alignment between system recommendations and manual evaluations.
-
Cost Efficiency: Achieved comparable results to ChatGPT-based recommendations at a significantly lower cost due to the scalable embedding-based approach.
-
Extensive Testing: Processed over 1,500 applications for 664 candidates and 287 vacancies, highlighting the system’s ability to handle large-scale operations.