As a defense contractor, it isn’t always easy to share information about what Praxis does. This is a particular problem when recruiting college students who want some insight into the kinds of work they’ll be doing. That’s why Praxis makes a conscious effort to develop intern programs that are just as interesting and mission-relevant as contract work.
To give you a good idea of some of the cool things Praxis does, here are two examples of intern projects. You’ll see that they are directly tied to technologies and cybersecurity happenings that you’ve seen in recent news. This gives interns a sense of how their work directly affects more local problems like business productivity, and as far-reaching issues in cybersecurity and national security.
VAST (Video Analytics with Speech & Text)
Praxis interns are tasked with using AI to extract key information from videos – information that would allow a user to search for videos without having to watch them. They provided the group with over 700 cell phone videos (~10 hours), recorded during the January 6 Capitol Riot. The team focused on different video aspects: audio, images, and metadata.
Using audio extracted from each video, the team was able to separate audio signals into specific voices (if multiple people were talking), build sound detectors to recognize specific noises (e.g., gunshots), and create text transcripts of spoken words. With the visual aspects of the video, they built a keyframe detector to identify unique still frames, used Optical Character Recognition (OCR) technology to extract text from signs, and leveraged Object Detection to identify interesting objects and approximate headcounts in still frames.
Video metadata was useful in providing information context such as GPS coordinates and creation dates and times. All of this extracted information was stored in a searchable data store, which enabled more complex analysis. For example, the textual content was analyzed to identify key terms and phrases that were prominent in the videos, and to get a sense of the overall sentiment in the video.
Additionally, a user could search against the stored data to find specific videos that fit specific criteria – for example, “Find me videos with less than 10 people and signs that contain the word ‘fight’.” As researchers in 2024 are exploring methods to create Multimodal AI models, they are happy to highlight this work as a precursor to that!
WISH (Wireless Information Survey Helper)
What value can an observer extract from the wireless signals emanating from their mobile device? Praxis interns collected sample data by intercepting packets sent by mobile devices to a designated router. Using this data, they were able to show the value of various analysis techniques by creating two applications: CometGo and CometLo.
CometGo identifies unique devices within a specific radius (such as an office building), and tracks when those devices enter or leave the location. Although the team identified a limitation in being able to consistently identify the same device from session-to-session (due to the prevalence of randomized MAC addresses), they were able to find possible workarounds that could uniquely identify a device.
CometLo identifies the specific location of devices by combining the packet information collected using multiple wireless receivers, along with their associated signal strengths to ascertain a location with a high level of confidence. Overall, this project showed how easy it is to create a pattern of life analysis just by collecting the packets being sent by a mobile device!
Praxis is already planning an exciting project for this summer’s interns that combines IoT development and machine learning. Stay tuned to hear how their interns develop creative solutions for the problems thrown at them!
Contact Praxis to be part of their team of engineers working on exciting projects.
Browse open positions..
SPONSORED CONTENT: This content is written on or behalf of our Sponsor.
