Machine Learning Research Intern
PACS AI Group · Medical imaging research · Boston & remote
Physics, machine learning, and research software.
I’m a physics student at UC Santa Barbara. I build software for medical imaging, autonomous navigation, astronomy, and quantitative research.
About
My work covers CT imaging, astronomy, embedded navigation, and live data systems. I handle the research code, evaluation, and user-facing software.
Outside the lab, I founded Borrow the Sky, a library astronomy-access program, and ran public stargazing nights for young people. At Phillips Exeter Academy, I co-led the Astronomy Club and turned an online forum into a shared satellite-data project.
Experience
PACS AI Group · Medical imaging research · Boston & remote
Summer Science Program · Astrometry and orbit determination · Georgia
UCSF Radiology & Biomedical Imaging · CT reconstruction and segmentation
Selected work
Astrometry, medical imaging, robotics, standalone VR, and Earth observation.
I helped collect three astrometric observations, recovered the asteroid’s heliocentric state with Gauss’s method, and propagated measurement uncertainty into all six orbital elements with Monte Carlo trials.
The measured astrometry was published in Minor Planet Circular 186587 under R. K. Vaid for object 263976.
At PACS AI Group, I built λ-AttentionNet, a multimodal model that combines TransUNet imaging features with CLIP vision-language outputs to draft radiology reports and express confidence per prediction.
I also built the PyTorch data pipeline: voxel normalization, anatomical segmentation, and OCR that parsed reports into seven structured sections.
I designed an insect-inspired familiarity circuit for panoramic navigation on a microcontroller. Procedural Kenyon-cell connectivity eliminates the projection lookup table.
Accepted as a spotlight and poster at the IROS 2026 Workshop on Insect-Scale Autonomy. The repository includes the implementation, simulation, and memory audit.
I’m a main dev on the native Quest and PICO ports, including the unified launcher, retail-content import, update path, and save-preserving migration tools.
I’m a main dev on the standalone Half-Life: Alyx port, with work across native runtime bridges, rendering, controller tracking, and Android packaging.
I co-built a local-first geospatial explorer for studying migration, urban growth, and activity change through satellite-derived proxy metrics.
Earth Engine computes the tiles. FastAPI, SQLite, React, and Leaflet handle time-series analysis, regional comparisons, animations, and exports.
Publications and workshops
NeurIPS 2026 · NEmo Workshop · accepted poster
Tests whether readable symbolic predicates in embodied language agents causally influence the next action.
IROS 2026 · Workshop on Insect-Scale Autonomy · accepted spotlight + poster
Tests quantized familiarity navigation under a 3.2 kB state budget with closed-loop LiDAR trials in a Seville ant habitat.
CoRL 2026 · SPAIS Workshop · submitted
Audits 36 evaluator units across six manipulation-safety benchmarks, separating simulator-internal labels from signals a runtime robot monitor could observe.
AIxVR 2026 · submitted
A framework for studying hidden capability asymmetries when AI-augmented XR enters face-to-face interaction.
NeSyDebates 2026 · submitted
Measures how in-context symbolic aliases change the persistence of intermediate arithmetic states across transformer layers.
ICWSM · submitted
Establishes finite-sample limits for rare-harm audits and evaluates prediction-assisted prevalence bounds that require fewer human labels.
IEEE ICASSP 2027 · submitted
Derives the weight-space modal tangent kernel for bilinear Fourier PINNs and tests its training-trajectory predictions against gradient descent.
Archive
Observing archive
Contact
I’m looking for research and engineering roles in physics, machine learning, and quantitative systems.