FgenXAI: A Generative AI Framework For Explainable Financial Records Summarization

Rakshit Rao, Manoj Mangam, Shivam Arora, Raahul Nallasamy, Sherin Bharathiy M, Aakarsh Malhotra, Alok Mani Singh

Keywords: RAG, Financial Records, Explainable AI, Generative AI, Jail Break Dataset, Agents

Abstract: Artificial Intelligence (AI) systems are widely used in various domains. These systems must be interpretable and explainable for end users. Techniques, such as SHAP and LIME are commonly used for this purpose. However, they often lack interactive explanations for end users with limited domain knowledge. To address this issue, we propose the FgenXAI framework that leverages generative and explainable AI to address the transparency and interpretability challenges posed by financial AI models. FgenXAI comprises four components: user query filtering, query parsing and context preparation, response synthesis, and response checking. We conducted an extensive study of the hallucination, refusal, and jailbreak properties of FgenXAI to showcase its efficacy. The FgenXAI framework reported an accuracy of 99% and a true refusal of 99% on an average on domain-specific self-curated datasets. Moreover, to check the reliability of FgenXAI, we curate a finance-specific jailbreak prompt database with 9,490 prompts, showcasing that FgenXAI is 95% immune.

Submission Number: 33