Selected coverage and applied projects in visual intelligence and multimodal AI.
My applied work connects multimodal foundation models, embodied reasoning, 3-D reconstruction, and public-safety sensing in settings where evidence is noisy, incomplete, and high-stakes. These projects translate uncertainty-aware reasoning, spatial understanding, and deployable decision support into trustworthy AI systems for public-interest analysis, mobility, secure deployment, and safety-critical environments.
Some of this work supported The Washington Post investigations included in its 2022 Pulitzer Prize for Public Service-winning coverage.
Supported crowd-density and flow analysis for The Washington Post's January 6 Capitol attack reconstruction, helping align visual evidence with radio traffic, dispatch records, movement patterns, and event timing.
Methods: crowd counting, density estimation, spatial registration, temporal alignment.
Role: Listed second among experts in The Washington Post Visual Forensics credits.
Assisted with multi-camera crowd-density reconstruction and spatial alignment for venue-safety analysis. The work helped compare local crowd conditions across footage and ground-plane views.
Methods: camera alignment, ground-plane mapping, density estimation, synchronized review.
Role: Lead for CMU analysis credited in the Washington Post video.
CMU LTI: Reported the lab's Astroworld analysis for The Washington Post.
View coverage Visual Forensics video Credit timestamp CMU LTI report
Contributed audio-visual analysis for timing, localization, and sequence reconstruction. The work helped compare citizen videos under camera motion, occlusion, and incomplete viewpoints.
Methods: event timing, temporal alignment, localization, audio-visual checks.
Role: Team member under Prof. Alex Hauptmann's supervision.
Supported reasoning about gunshot timing, camera geometry, trajectories, and location bounds. The analysis helped connect visual evidence with scene constraints under uncertainty.
Methods: audio-visual sync, camera geometry, trajectory analysis, localization.
Role: Team member under Prof. Alex Hauptmann's supervision.
Used 3-D reconstruction and geometry-aware modeling to reason about scene layout and shooter positions. The work supported line-of-sight analysis from visual evidence in a contested public-space event.
This case illustrates how spatial reasoning, camera geometry, and interpretable visual evidence analysis can support human-centered investigation when direct observations are incomplete or conflicting.
Methods: 3-D modeling, camera geometry, spatial reasoning, visual evidence.
Role: Team member under Prof. Alex Hauptmann's supervision.
Presented an early public-safety prototype using smartphone videos for event localization. The system explored how multiple consumer videos could support emergency situational awareness.
Methods: multimodal sensing, video calibration, audio-visual localization.
Role: Team member under Prof. Alex Hauptmann's supervision.
Covered the prototype's technical foundations, including calibration, robustness, and deployment constraints. The work focused on localization from distributed video sources.
Methods: video calibration, multi-view alignment, event localization.
Role: Team member under Prof. Alex Hauptmann's supervision.
Provided technical perspective on metadata, archival sourcing, and responsible verification. The work supported careful interpretation of sensitive visual materials and provenance evidence.
Methods: provenance analysis, metadata review, privacy-aware handling.
Role: Team member under Prof. Alex Hauptmann's supervision.
Note. These projects involve high-stakes settings. Analysis supports, rather than replaces, journalistic verification, legal review, expertise, and human judgment.