Animal Identification
Interested in identifying animals from camera trap footage? I developed a computer vision pipeline that automates wildlife detection, segmentation, classification, and metadata extraction from raw videos, turning days of manual review into structured datasets ready for analysis. By rapidly detecting species and pairing behavioral data with time/location context, this system opens new avenues for studying animal behavior and supports data-driven conservation and policy decisions. New classes can be trained in under 6.5 hours, making it adaptable for diverse ecosystems and research goals. Contact me if you’d be interested in training your own footage or implementing this system for your projects.
Monthly Webimars
Join my monthly webinars on the second Saturday of every month. I discuss my methodology and process of scaling this project to different species. Have more questions? Send me an email, and I'd love to answer them further.
Other Projects
All my projects are impact-driven. I led the design of a voice-controlled assistive device for Ken, an ALS patient, to support vertical arm movement. Our team evaluated user needs, compared mechanisms, and settled on a belt-driven design that distributed forces comfortably. Iterative testing revealed torque limits, so we added countersprung tension bands, and a design review with Arthrex engineers introduced a dogbone constraint that eliminated gear skipping. Programming the interface to respond to “JARVIS” personalized the device for Ken; the final prototype reliably assisted daily tasks. I tested a carbon spool linear extension to explore high-speed actuation for robotics. It achieved very fast extension, but was only rigid in two planes and couldn’t support larger loads. This led me to explore box slides and cascading lift systems. I also designed and built JBOT, a <$40 Arduino robotics kit to make hands-on STEM accessible. I field-tested it at four summer camps, impacting 130+ students, and iterated the hardware and instructions based on observed failure modes. The final kit was robust enough for students to build, debug, and complete projects independently.
