Talk to Your Brain: Artificial Personalized Intelligence for Emotionally Adaptive AI Interactions

The process of =making an Artificial intelligent system to replicating human emotional understanding and adapt its responses contextually by tailoring its responses based on individual cognitive and emotional states is called Artificial Personalized Intelligence (API). In this paper, we present a finetuned emotionally adaptive AI pipeline capable of generating personalized, human-like responses. Using a custom dataset based on six universal emotions—Sadness, Happiness, Fear, Anger, Surprise, and Disgust—collected from interviews with 30 participants, we have explored two approaches: fine-tuning a LLaMA-3 8B model with the Low-Rank Adapter (LoRA) technique and employing a Retrieval-Augmented Generation (RAG) agent-based framework over the same LLaMA-3 8B model. Emotion classification, prompt engineering, and model fine-tuning were used over the dataset to capture emotional and personalized subtleties in the responses based on different individuals. We conducted a comprehensive analysis, evaluating performance across candidates, emotions, and overall performance, using metrics such as Mean Squared Error (MSE) and Pearson Correlation Coefficient to measure the difference among the outputs from both pipelines and actual human emotional responses. Our results show that prompt engineering combined with LoRA-based fine-tuning significantly enhances the ability to engage in emotionally intelligent and personalized conversations, whereas the RAG-based approach underperformed due to the pretrained LLAMA-3 model’s restrictive neutral and adversarial training. This highlights that fine-tuned LLMs are effective in replicating human emotions for personalized intelligence.