2024 LAS – NC State Computer Science Senior Design Teams 

Since 2016, researchers at the Laboratory for Analytic Sciences (LAS) have sponsored approximately 25 senior design projects through the NC State Computer Science Senior Design Center, including four in 2024. Through this engagement, the LAS has worked with well over 100 students on impactful, technical R&D projects.

The idea of the Senior Design course is to offer students a hands-on learning experience working on real-life problems sponsored by area businesses and research organizations (the LAS being one of them). Students work in small teams under the guidance/mentorship of representatives from the sponsoring organization throughout the term to develop solutions.

Over the years, projects have spanned a wide variety of technical topics, from cloud computing efficiency upgrades to gamifying the task of machine-learning annotation. All project proposals stem from mission needs in analytic technology and tradecraft from LAS stakeholders, and are developed over the months preceding each semester. After the semester concludes, LAS staff work to transition positive results onto/into mission systems. Significant, tangible successes have resulted. Below we will briefly describe the four projects undertaken this year.

Note: A full accounting of these four projects, or any of the other Senior Design projects that the LAS has sponsored in the past, is beyond the scope of this article. For more detail on any specific current or past project, please reach out to the Contact information provided at the LAS website. You may also refer to previously published articles such as this article we wrote for the LAS 2022 Symposium which offers basic descriptions for most LAS Senior Design projects up to that point.

Video Content Management System (VCMS)

The VCMS team prototyped how we could improve visibility into the EYERECKON pipeline (a pre-existing video processing system developed at the LAS). Specifically, the team was asked to expose the individual components that make up our pipeline and the corresponding running status for each video. As a result of their code, at ingest a user can customize the pipeline to their liking. The students provided options to select specific preprocessing steps, sampling logic, frame (and object) embedding model(s), and specific object and event detection models. In addition to being configured with a simple API endpoint, these choices were also mapped verbatim to backend processing logic to ensure that the UI and corresponding logic could grow and adapt to our future needs.

A photo of the VCMS Senior Design team presenting their work.

LOADWOLF

The LOADWOLF team took on a very challenging R&D problem of building a custom scheduling system for cloud compute analytics. The purpose is to offer system owners the capability to define their own desired system load (i.e. amount of total cloud resources utilized at any given time). The LOADWOLF scheduler will then generate a schedule that should produce an actual system load similar to that which the owner desired. Efficiency gains and system stability are the expected outcomes of this scheduling capability. The team also developed a user interface and performed an initial round of testing on the scheduler. Results on a relatively small test cloud, with a relatively small set of analytics, are very positive. The actual load curve on the actual system did indeed appear to match that which was defined as the desired load with good accuracy. On the strength of this outcome, the next steps will be to scale the solution to a much larger cluster, with many more analytics. We’re anticipating that a future Senior Design will take on this challenge!

A photo of the LOADWOLF Senior Design team presenting their work.

AI Artifact Management System

The technical world is producing huge numbers and versions of data sets and AI models. There are some 350,000 models on Huggingface for example. This number is likely to continue exploding in the coming years, so most organizations need some form of approval and tracking status for their own purposes regarding which models and datasets they are using. Each model/dataset may have differing licenses, technical requirements, applications, etc. To assist an office tasked with adjudicating and tracking the status of all such models under use, or consideration for use, in their organization, this Senior Design team designed and developed a custom web application specifically for that purpose. The application is fully functional, and we expect to deliver a live version to at least one stakeholder by year’s end. Below is a simple screenshot of the Artifact List page.

A screenshot of the application showing a table of AI artifacts and their statuses.

Instrumenting STT

In this project, the Senior Design team enhanced the Common Analytic Platform (CAP), an LAS-developed research platform, by integrating granular instrumentation into its workflow. The team focused on capturing detailed metrics of how analysts interact with speech-to-text (STT) data, including time spent on key tasks such as reviewing search results or jumping to keywords. The team created a custom javascript instrumentation library that can be incorporated into an existing STT application like CAP enabling collection of the instrumentation data. The data is then stored in a SQL-based database for subsequent visualization and analysis.

Analysis of the data is expected to offer valuable insights on workflow efficiencies and how analysts use STT data in search and discovery. The LAS team plans to explore transitioning these capabilities to the corporate network in 2025 so that this data can be used to inform agency decisions on whether to invest in improving STT algorithms, use different ST algorithms for specific use cases, or optimize user interfaces with new technologies.

A system diagram showing the various components of the STT Instrumenter

This material is based upon work done, in whole or in part, in coordination with the Department of Defense (DoD). Any opinions, findings, conclusions, or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the DoD and/or any agency or entity of the United States Government.