The recent advances of Artificial Intelligence (AI) has tremendous potential in enhancing existing learning applications. Many existing platforms have components to assess learners’ performance; however, very limited work exists for Sanskrit learners, especially systems which help to correct pronunciation mistakes. Since Sanskrit, an ancient Indian language with precise phonetic rules, is very challenging to learn without the mentorship of an experienced tutor. Hence, learning systems intended to teach Sanskrit need the emotional intelligence of a human guide. In any learning system, the commitment, zeal, and motivation of the learner play a major role. A system that teaches, assesses, and corrects pronunciation errors of Sanskrit learners needs affective personalized pronunciation feedback to encourage and motivate learners. Hence, in this work, we design a Sanskrit Voice Bot integrated with emotional intelligence to provide personalized affective feedback. This paper compares the performance of Gemini 2.5 flash and Llama 3.2 models in producing pronunciation feedback for users. The experimental results showed an average feedback quality of 82.7, scored by Llama 3.2 with 9.3 variance, and an average score of 75.4 by Gemini 2.5 flash with 12.1 variance. These results signify improvements in both actionability and affective positive tone for Llama 3.2.