Surveying world-action models for robot manipulation, geometric and latent-space, covering diffusion policies, vision-language-action models, and robot self-improvement from human video, and mapping the benchmark landscape for world-action model evaluation to ground a joint research proposal.
Supervised and semi-supervised pose-estimation training and inference on multi-view mouse video data using Lightning Pose (PyTorch), running on CUDA and Linux.
Built an end-to-end pipeline converting raw financial PDFs into bank-ready compliance spreads, with an anti-hallucination validation layer, derived-metric computation, multi-format deliverables, and deployment packaging.
Built a responsive dashboard UI for visualizing ANN-generated research data, including a reusable filtering component and rendering performance work.
Generative computer vision research on synthesizing rare astronomical images to address data scarcity in astronomical imaging datasets.
Converting raw hand-drawn concept maps into a structured, ANN-ready dataset and extracting graph-theoretic features from them.
Evaluating GPT-3.5 and GPT-4 for data-analysis capability on large tabular datasets, and characterizing their failure modes.
Volunteer across two organizations in the Houston area, both rooted in the South Indian cultural community. A month of service work at Shirdi Sai Jalaram Mandir in Sugar Land, and the technical and photographic side of Sai Meghna Dance School, a Kuchipudi classical dance academy run by my mother for over twenty-five years.
Co-founded and led the organization, directing global-issue awareness campaigns, donation drives, and fundraising sports events.