The digital finance sector is increasingly demanding intelligent solutions to help retail investors navigate the complex in formation of the stock market. One of the key challenges is delivering personalized investment recommendations that align with individual goals while maintaining a natural conversational experience. To address this, we propose Reksata, a Trans former based Conversational Recommender System (CRS) for personalized digital stock recommendations. Reksata leverages a f ine-tuned Large Language Model (LLM) based on GPT-3.5 turbo, optimized using domain-specific financial data over three training epochs, achieving a training loss of 0.362. Evaluation results show that the GPT-based model outperforms a fine-tuned RoBERTa baseline in semantic similarity tasks, achieving a precision of 0.9565, recall of 0.9396, F1-Score of 0.9479, and cosine similarity of 0.7920, compared to RoBERTa’s F1 Score of 0.8893 and cosine similarity of 0.6683. The system supports multi-turn dialogues, interprets and adapts user preferences, and dynamically refines recommendations based on conversational feedback. These findings highlight the potential of Transformer based CRS to provide more responsive, context-aware, and user-focused stock investment guidance within digital finance platforms.