AI research · Visualization · Human understanding
Understanding AI
to make it safer.
I’m Shivam Raval, an AI research scientist exploring how we can better understand, visualize, and interact with machine learning.
A little context
I'm a research scientist focusing on making machine learning more interpretable through visualization and interactive systems. My work combines techniques from deep learning, human-computer interaction, and data visualization.
My research aims to bridge the gap between powerful ML models and human understanding of model internals. This involves explaining and visualizing clustering structures in high-dimensional data and interpreting latent activations in frontier AI models. I also use interactive visualizations to create explanatory articles that make AI interpretability methods more accessible.
Ideas into interfaces
Selected work
Visualization / IEEE VIS 2024
Hypertrix: An Indicatrix for High-Dimensional Visualizations
Making the distortions in high-dimensional projections visible.
Best Short Paper AwardInterpretability / Google PAIR
Mapping LLMs with Sparse Autoencoders
An interactive exploration of the features inside language models.
The latest
Recent publications
AI Agents and the Future of VIS
Workshop proposal · arXiv 2026
Riemannian-Manifold Steering: Geometry-Aware Generative Autoencoders for Label-Free Steering
arXiv preprint · 2026
Dissociating Decodability and Causal Use in Bracket-Sequence Transformers
arXiv preprint · 2026
H-Probes: Extracting Hierarchical Structures From Latent Representations of Language Models
arXiv preprint · 2026
Keep the conversation going
Curiosity is better shared.
Research, ideas, or something interesting you’re building.
sraval@g.harvard.edu