-
Boeing Research & Technology
Kyle DeProw
Biography
I'm a Saint Louis University M.S. Engineering graduate (M.S. in Dec 2022) with a weird love for Robotics and AI that started all the way back in my M.E. undergrad at SIUE (B.S. in May 2016). My background has been heavily focused in the field of algorithm development for Artificial Intelligence, Machine Learning, and Deep-Learning applications primarily applied to the domains of Generative Modeling, Multivariate Time-Series Prediction, Anomaly Detection, and Adversarial Machine Learning. I've also been heavily engaged personally in the Natural Language Processing space ever since my first Seq2Seq lecture in university (but please don't yell at me if I just don't agree that ChatGPT is the solution for everything!). My extensive time in this R&D, cutting-edge, fly-by-the-seat-of-your-pants, environment has also really highlighted the value of championing proper software development and culture through Microservice software design, containerization, and cloud-computing. Unfortunately, this runs the risk of developing some quirks, such as my vim elitism, spaces >>> tabs, and insistence on clean, bite-sized, well documented git commits. I'm curently working as a AI/ML Research PI with Boeing Research & Technology where I specialize in Real-time Anomaly Detection using DNN architectures through behavioral modeling leveraging Bayesian strategies to determine deviations in expected state-space conditions. When I'm not working, you're probably likely to find me spending time with my pup Dani, hiking, consuming copious amounts of Brandon Sanderson/Niel Gaiman.....or well coding (I know I'm boring!). Read on and follow me on my social media below!
My Resume
Email me at kydepro@gmail.com
if you have any questions!
Click here to download my resume.
Technical Skills
Fluent languages
- Python
- C++
Frameworks
- tensorflow/pytorch
- pandas
- cmake
- docker
- kubernetes
- git
Publications
Patent Disclosure
Work Experience
Senior Machine Learning Researcher
Dec 2021 - Current | Berkeley, MO
- Generative Modeling for Anomaly Detection: Led a $2.25M Generative Modeling research effort tasked with bolstering platform cybersecurity through Time-series, Multivariate Prediction and Anomaly Detection utilizing Sensor-Fusion fed Bayesian DNN structures to detect operational deviations. My success in this field was critical in winning customer capture opportunities that cumulatively returned 7x ROI
- Ensemble Modeling with NLP: I wrote, proposed, and led a team of three engineers to execute a research effort focused on augmenting the performance of PredAI aviation anomaly classifiers by ensembling NLP classifiers that ingested software flight logs from commercial aircraft. Developed a tech stack based on LogRobust, Drain, and PyTorch that improved our anomaly detection F1 scores ∼5-10%
- Adversarial Machine Learning: Spearheaded the Exploitable AI and Adversarial Machine Learning portion of BR&T’s Product Security’s realignment. I focused on researching, proposing, and maturing scalable and deployable strategies to mitigate the effects of backdoor training data poisoning for aviation which, once prototyped, resulted in a novel research derivative and $200k of funding.
- Proposal Writing and Funding Capture: Utilized AI/ML subject matter expertise to identify opportunities for designing and authoring competitive business discriminators in response to customer project announcements culminating in a total of $17M in captured funding and another $10M in potential future projects.
- Mentoring and Knowledge Sharing: Formally mentored a team of five early-career engineers in ML/AI, Software Development, and Proposal Writing. This involved leading a hands-on, brown bag ML workshop bi-monthly to discuss state-of-the-art, solidify fundamentals, promote cross-team communication, and eliminate knowledge silos amongst the team.
-
Boeing Defense, Space, & Security
AI and Autonomy Engineer
Dec 2020 - Dec 2021 | Hazelwood, MO
- Reinforcement Learning: Architected and Implemented DQN, A2C, and TD3 Reinforcement Learning algorithms to solve path-planning and refueling missions and encode this learning into a general solution. Optimal solution finding accelerated mission planning exercises by at least a factor of 5 in laboratory settings as compared to traditional analyst approaches.
- Autonomy and AI: Designed and implemented traditional AI agents to execute automated behaviors such as ”search and rejoin”, ”wingman follow”, and ”reroute around area” in the Aerospace Simulation environment AFSIM. Solutions involved algorithms spanning Rules-based, Optimal Search (A*), and Finsite State Machines.
- Data Pipelining and Experimentation: Developed methodologies to interact with AFSIM simulation data to effectively create a high-fidelity, robust AI gym capable of supporting sophisticated experimentation with 100’s of platforms in wargaming scenarios. Designed mechanisms to transfer, consolidate, and archive the large amounts of data utilizing common machine learning optimization techniques, like quantization, to reduce the resource consumption and minimize latency.
-
Boeing Global Services
Software Engineer - Simulation
Feb 2019 - Dec 2020 | Berkeley, MO
- Real-time Simulation: Developed capabilities for real-time RHEL Linux training platforms specializing in IFF, ILS, Weapons, Electronic Warfare/Cybersecurity, and Radar functionality. This work spanned over two different platforms, each with technology stacks ranging multiple languages (primarily C++ and Python) with ¿50k LoC.
- Constructive Simulation: Primary development lead for integrating maritime simulation capabilities into the industry standard simulation tool AFSIM. Through this effort, I expanded the operational domain of this software by 33% and enabled novel experimentation that was published at BTEC.
- Standard Aircraft Protocols: Experience with aircraft protocols including MIL-1553-STD, Link-16, ARC-231, and RS-232 protocols.
- DevSecOps and CI/CD: Implemented containerized Docker solutions to enable AWS cloud capabilities for a variety of enterprise use-cases. Championed proper software development and design practices (Git, Jenkins, Jira).
-
Saint Louis University
Robotics and AI Researcher
Jan 2017 - Feb 2019 | St. Louis, MO
- Academic Research and Publications: Research grant funded position to lead research relevant to NSF Cyber-Human Systems programs which included fields such as Robotics, AI, and Machine Learning.
- Robotics: Designed, machined, and programmed a 5-DOF anthropomorphic arm manipulator for a two wheeled telerobot that was actuated and controlled to mimic the movement of an operator's arm.
- Perception Systems: Leveraged video data from a Microsoft Xbox Kinect to implement inverse kinematic solutions that solved operator arm pose angles transposed on a telerobotic arm.
- Supervised Learning: Implemented Tensorflow neural networks and other ML techniques to learn tactile features on robotic platforms in the absence of visual validation.
- Reinforcement Learning: Explored using Reinforcement Learning algorithms to control the telerobotic platform using rewards to guide the robot's state rather than much costlier alternatives found in Modern Control techniques.
-
Dynamic Controls