AI/ML Engineer (Computer Vision)

CACI•Ashburn, VA

About The Position

CACI is seeking an AI/ML Computer Vision Engineer with Agile methodology experience to join its BEAGLE Agile Solution Factory team supporting the Customs and Border Protection client in Northern Virginia. The engineer will design, develop, and implement custom software solutions for AI/ML pilot projects and use cases, collaborating with prototyping teams to deliver working applications on short timelines. The role involves solving complex technical problems, participating in Agile delivery activities, and staying current with AI technologies and practices. The engineer will also help build scalable, highly available systems and work with data scientists to integrate applications with AI/ML, big data, and business intelligence solutions. The position focuses on computer vision for CBP full-motion video feeds. Responsibilities include designing and training custom YOLO architectures, architecting secure video ingestion and preprocessing pipelines, defining data annotation and curation strategies, addressing data challenges such as class imbalance and variable conditions, and optimizing inference performance and accuracy using hardware-acceleration technologies. The engineer will collaborate with video streaming engineers, software engineers, and system architects to integrate live computer-vision inference into the broader application. The role requires the ability to work independently on complex tasks with limited direction.

Requirements

U.S. citizenship and ability to pass a CBP background investigation, including checks for felony convictions, illegal drug use, and misconduct such as theft or fraud.
Bachelor’s, master’s, or Ph.D. in Computer Science, Data Science, Electrical Engineering, or a related field with an AI/ML focus.
At least 7 years of related technical experience, including at least 4 years developing and deploying deep-learning/computer-vision models, with a strong focus on object detection.
Experience with deep-learning frameworks such as PyTorch (preferred) or TensorFlow, and computer-vision libraries such as OpenCV.
Advanced implementation and fine-tuning of YOLO models, including Ultralytics YOLOv8/v9/v10 and custom YOLO backbones.
Experience training deep-learning models from scratch, including custom anchor-box design, loss-function modification, and hyperparameter tuning.
Experience with GPU-accelerated computing using NVIDIA CUDA, cuDNN, TensorRT, or ONNX.
Experience working with video streams (RTSP, HLS, or files) and frame-by-frame processing.
Expert-level Python programming and ability to write clean, modular, optimized code.
Experience with data curation, active learning, and annotation tools such as CVAT, LabelStudio, or Roboflow.
Agile methodology experience.

Nice To Haves

Experience with multi-object tracking algorithms such as DeepSORT or ByteTrack.
Familiarity with cloud-based AI infrastructure such as AWS SageMaker or EC2 GPU instances.
Experience deploying models to edge devices such as NVIDIA Jetson or tactical hardware.
Prior experience working with federal agency datasets or Customs and Border Protection.

Responsibilities

Design, develop, and implement custom software solutions for AI/ML pilot projects and use cases.
Collaborate with prototyping teams to deliver working software applications on short timelines.
Analyze complex project-related problems and develop innovative solutions involving technology, methodology, tools, and solution components.
Participate in Agile delivery phases and ceremonies, including release and sprint planning, artifact creation, sprint testing, demonstrations, and retrospectives.
Stay current with advancements in AI technologies, tools, and best practices.
Design systems that scale horizontally and support high availability.
Work with data scientist teams to integrate applications with AI/ML, big data, or business intelligence solutions.
Work independently on complex tasks with little direction or management oversight.
Design, modify, and train custom YOLO architectures optimized for CBP full-motion video feeds and their environment, resolutions, and challenges.
Help architect a pipeline to ingest, clean, and preprocess production-recorded video into a secure training and sandbox environment.
Define annotation and labeling guidelines, manage dataset curation, and develop strategies for class imbalance, varying weather, and low-contrast conditions.
Optimize model inference speed and accuracy using hardware-acceleration toolkits such as NVIDIA TensorRT, DeepStream, and CUDA.
Collaborate with video streaming engineers, software engineers, and system architects to integrate live computer-vision inference with the broader application.