Working at the intersection of explainable AI (XAI), AI alignment, and safety in high-stakes domains—focusing on interpreting black-box models, evaluating the reliability of XAI methods, and developing in-house foundation models for tabular data, particularly for fraud detection and other mission-critical applications. Contributed to the development of the DLBacktrace explainability method and co-created a benchmarking library for evaluating state-of-the-art XAI techniques. Explored model-agnostic AI alignment and optimization using post-hoc explainability methods, with experimental work across a range of architectures including CNNs (VGG-19), BERT, and Llama models, spanning diverse modalities and tasks. Led and mentored interns, played an active role in recruitment, and worked on proof-of-concept (POC) initiatives aimed at advancing internal algorithmic capabilities. Represented the R&D team in client engagements and at industry events, including the 5th MLOps Conference, to present and demonstrate AryaXAI’s integrated AI solutions. AryaXAI Alignment Labs, AryaXAI.com (Arya.ai [Lithasa Technology Private Limited], an Aurionpro Company).