During my time at Neuligent, I designed AI-based psychometric analysis tools. To do this, I processed a database containing discussions of people and their HEXACO profiles, and trained models to predict psychometric trends. HEXACO is the name of the reference psychometric test used in my work. In this page I'll briefly present the proofs of concept I've produced using psychometrics.

This interface lets you import text and select which HEXACO category to examine (HEXACO is a psychomtric test evaluating Humility, Emotionality, eXtraversion, Agreableness, Conscientiousness, and Openness). In this case, Openness has been selected so we see in blue which parts of the text show signs of openness, and if there was any, we would see in red signs of anti-correlation with openness:

Here we can simulate an interview thanks to a chatbot I created using a LLM, at the right of the screen we can analyze the psychometric results of the interview to have an idea of the candidate’s personality:

In this interface, we can select a job description, then create a HEXACO-based personality test adapted to the job. Contrary to a non-personalized test, this will be far more relevant to know more about a candidate. Questions are generated by AI, and the user (HR manager) must select which ones to keep.

And here, we can simulate a user completing the test and see their results: HEXACO measurements added to comments about their personality and work habits.

This training environment enables us to configure chatbots to simulate a situation (customer service, a person in distress, telephone sales, etc.). We can configure the situation, the bot's personality, and the user's evaluation elements. Speech-to-text and text-to-speech have been implemented to make the situation more realistic, and the bot has been trained using data from real conversations to appear natural. In this example, I've created a situation involving an Internet outage: an irate customer calls customer service, and the user has to respond to the customer's concerns and direct him towards a solution. At the end of the discussion, we can obtain the user's strengths and weaknesses, as well as suggestions for improvement.