Resume

Dissertation on visualizing and interpreting high-dimensional data: theory and applications.

• Mathematical & engineering principles for training foundation models
• Geometric deep learning and manifold learning techniques
• Human-computer interaction and visualization

Major in Physics with a focus on computational modelling, simulation, and experimental research.

• Computational physics modelling and simulation in Python, MATLAB, and C++

Developing novel LLM inference optimization techniques and integrating them into client-facing products.

• Reduced LLM API inference cost by 30% through novel optimization techniques
• Curveball Steering: a curvature-aligned method for steering LLM behavior (arxiv.org/abs/2603.09313)
• Developing robust domain-specific AI personas (arxiv.org/pdf/2510.22170)

Designed and curated high-quality training datasets for foundational models in advanced mathematics and physics.

• Implemented rigorous manual evaluation processes to ensure dataset quality and model performance

• Engineered production-ready interactive platform for LLM interpretability using React and PyTorch, focused on Sparse Autoencoders (SAEs): pair.withgoogle.com/explorables/sae/
• Developed research direction and initial analyses on novel SAE phenomena like shrinkage and feature splitting

• Studied interpretability and sensemaking challenges of high-dimensional embedding visualizations in text corpora and image datasets
• Developed novel approaches to inspect projection distortions, steer embeddings by user-defined concepts, and explain cluster structure
• 3 conference presentations (IEEE VIS 2023–24, CHI 2026), 3 workshop presentations (NeurIPS 2023, ICLR 2025–26), 3 articles under review

• Led cross-functional teams of 6+ engineers and researchers, establishing development workflows, conducting code reviews, and managing Git repositories
• 3 successful publications in ML conference venues and new ongoing research

• Mentored undergraduate and early graduate students on explainable AI, visualization, and LLM interpretability
• Three accepted workshop submissions and four papers in preparation

• Organised and led student activities and outdoor trips aimed at improving mental and physical health

Frontiers in NeuroAI, Kempner Institute Symposium

Presenting research on understanding and interpreting large-scale activations in language models during reasoning tasks.

IEEE VIS 2024

Best short paper award winner on novel techniques for visualizing and identifying anomalous distortion in visual projections of high-dimensional data.