Hello!
I'm Aadya Agrawal, and this is my home on the internet.
Get in touch aadyaa3@illinois.edu
View my resumedrive link
I'm currently doing my master's in Computer Science at the University of Illinois at Urbana-Champaign, specializing in 3D motion prediction and generation using computer vision.
Before UIUC, I did my undergraduation in Mathematics and Computing at Indian Institute of Technology (IIT) Delhi. I worked on a variety of projects ranging from conditional generative modelling for 3D scene augmentation to exploring neurobiological learning processes through fMRI analysis. For more information, please check out my resume.
When I'm not in front of a computer screen, I'm probably watching sports, searching for new food spots, or crossing an item off my bucket list.
- Python
- C++
- SQL
- JavaScript
- HTML/CSS
- Java
- PyTorch
- TensorFlow
- React
- Node
- Express
- Bash
- Git
- MongoDB
- Jupyter
- CUDA
Developed a hierarchical attention-based diffusion transformer for multi-agent motion prediction that outperforms SOTA models on long-term predictions and close-proximity interactions. The model accurately predicts complex 3D human motion sequences, with predictions shown in color and ground truth in black.
Developed a novel clustering algorithm to identify activation patterns in fMRI signals during visual memory tasks. The visualization shows brain activation in two experimental conditions (left and center) and their differential response (right), revealing distinct activation (yellow) and deactivation (blue) patterns across brain regions.
Created a tool to infer relational graph structures from LiDAR point clouds for structured generative modeling of high-traffic driving scenes. Leveraged conditional-GANs to synthesize semantically valid objects and scenarios, generating diverse training data for autonomous vehicle perception systems.
Modeled ball-by-ball match outcomes using Monte Carlo simulation and hierarchical empirical Bayes methods, achieving an 8% improvement in predictive accuracy. Implemented LSTMs for time-series event forecasting and validated model reliability through betting-market alignment analysis.
Designed a contrastive learning framework for Android malware detection, achieving 94.8% accuracy on real-world datasets. Benchmarked the classifier's adversarial robustness and explainability to ensure enterprise-grade reliability and deployment readiness in production security systems.
Built a full-stack word association game using React and Flask to predict interpersonal compatibility from language patterns. Trained a Siamese neural network that achieved 90.9% recall and 83.3% F1 score on human-rated compatibility evaluations.
Developed a full-stack web application connecting users with nearby workout partners through intelligent matching based on sport preferences, schedules, and skill levels. Built with a REST API backend, Firebase authentication, and a responsive mobile-first interface.
I developed a scalable autoencoder utilizing hierarchical variational flow-matching for high-fidelity set generation. This project outperformed diffusion and flow baselines in terms of large-sample fidelity and diversity, and introduced an efficient, permutation-equivariant sampling pipeline without relying on domain-specific neural architectures.
I created a data-driven measure of defensive pressure using player tracking and event data from Eurocup 2020 for the Women in Sports Data Symposium in 2022. This work included developing intuitive pitch maps and heatmaps to visualize spatial trends and player behavior under pressure, and building a composite metric to evaluate pass quality based on accuracy, progression, and tactical impact.
I wrote a survey exploring the influence of social network structure and game theory on negotiation within multi-agent systems. The paper analyzed models of stubborn agents, cooperative value-sharing, and social strategies across evolving networks, proposing a strategic framework applicable to logistics, finance, and large-scale operations.