Artificial Intelligence for Human Resources

In recent years, Artificial Intelligence has transformed the way companies operate. The rise of Large Language Models (LLMs) has opened up new possibilities and new perspectives.

LLMs are Artificial Intelligence models trained on vast amounts of data that can understand and process natural language and generate text. These models are trained on huge volumes of text from the internet and can produce coherent, readable output. GPT-4 V, the architecture behind ChatGPT, is one of the most recent examples of this kind of innovation. These advanced technologies offer companies a range of distinct advantages, from personalizing the customer experience to reducing operating costs and automating repetitive tasks.

This article looks at a practical HR use case for generative models, showing why they can be a valuable choice for simplifying and significantly speeding up recruitment and selection processes.

What does HR need?

Recruiters often have to review a large number of CVs to identify the skills and characteristics that best match a specific role and company culture. This takes time and can also lead to mistakes, oversights or, in some cases, the introduction of bias, all of which can affect the overall quality of the selection process. To allow recruiters to focus on the most important aspects of hiring, AI models can help automate and improve both the effectiveness and efficiency of recruitment activities.

There are already software tools capable of extracting information such as age and educational background through tagging mechanisms. In many cases, however, these systems are not based on AI algorithms, and the results are not always satisfactory or comprehensive. Even among tools that explicitly claim to use machine learning models, performance indicators are often not disclosed, making it difficult to assess the actual quality of the results.

Case study: an application example

GPT models, developed by OpenAI, belong to the LLM family. These models are based on artificial neural networks that can process sequences of data efficiently and capture long-range relationships within a text. This makes them particularly suitable for tasks involving natural language processing.

In this context, GPT models were tested in an HR setting. More specifically, two different types of CVs were considered: narrative-style CVs and list-based CVs. The reason for this choice is simple: since CVs are written by different people and therefore vary widely in structure and style, the goal was to see how the models would perform across different text formats.

Following a selection process, one of the GPT models was chosen for its ability to handle a wide range of tasks with minimal prompting, including the in-depth understanding of a text and the generation of new content.

Entire CVs were then submitted to the model using a carefully designed combination of parameters, asking it to extract hard skills, soft skills, hobbies and personal details. The results were remarkably accurate and comprehensive, and not directly comparable to those obtained through keyword extraction methods such as KeyBERT or other open-source models, even though those approaches can still provide useful results.

It was also observed that the quality of the output did not change depending on the type of CV provided. Whether the input was a long narrative document or a bullet-point list, the model was able to extract the required information effectively.

For example, we created a sample CV (shown in Figure 1) and, without any preprocessing, submitted the PDF directly to the model. The algorithm was able to identify:

  • Hard skills: Microsoft Office, Windows, Ubuntu, C, C++, Python, SQL, Matlab, Wolfram, Mathematica, Spark, LaTeX
  • Soft skills: strong team spirit, ability to adapt to different environments, good communication skills
  • Hobbies: cycling, scuba diving
  • Personal information: Name: Mario Rossi, Date of birth: April 1, 1995, Address: Via Cavour 1, Milan, Phone number: 1234567890, Email: mariorossi@gmail.com
  • Languages: native Italian, advanced English, intermediate French

The quality of the output shows that the model not only provides accurate and relevant results—such as identifying hard and soft skills—but can do so without any prior data preprocessing, taking the CV directly as a PDF input.

This is not possible with earlier tools such as KeyBERT, an open-source library based on BERT embeddings and cosine similarity. KeyBERT requires text preprocessing: in addition to removing less relevant words, it is necessary to manually define the portion of text from which keywords should be extracted.

For example, the keywords extracted by KeyBERT from a paragraph on technical skills include terms such as knowledge, office, programming, Mathematica and languages.

The accuracy of LLM outputs enables the automated population of SQL and NoSQL databases, as well as data lakes.

However, these are not the only use cases for LLMs in HR. These tools can operate at different levels of complexity: they can be used not only for text search and extraction, but also for semantic analysis. For example, they can identify terms with positive or negative connotations, or detect language associated with specific traits such as ambition.

This type of analysis is particularly useful for roles that require specific characteristics. Take, for instance, a sales manager position, where ambitious candidates are often preferred. By asking the model to extract positively connoted terms from Mario Rossi’s CV, the output includes words such as passionate, creative, professional, cheerful, ambitious, dedicated, capable, adaptable and communicative.

What are the benefits for HR teams?

The tests and comparisons show that GPT models deliver accurate and comprehensive results quickly, with minimal effort required from the user. Compared to traditional solutions, these models provide higher-quality and more precise outputs. This makes them a valuable addition to recruitment workflows, helping streamline and improve the well-known process of turning raw data into information and, ultimately, into actionable knowledge.

LLMs can also support a wide range of HR activities beyond CV analysis. For example, they can be used to:

  • generate job descriptions
  • detect potential plagiarism when candidates reuse content from external sources
  • provide personalized recommendations to HR managers

Overall, these tools can represent a real competitive advantage, positioning organizations that adopt them within a more advanced and innovation-driven landscape.

If you’d like to explore this topic further, get in touch with us through the contact form