RESPONSIBILITIES

• Own the computer vision capability end to end — from problem framing and model selection through training, evaluation, and deployment on AWS
• Build and maintain data-gathering pipelines to assemble massive training sets — including writing scrapers and crawlers to source imagery at scale from across the open web
• Wrangle large, messy image datasets: cleaning, deduplication, filtering, normalization, and storage
• Build and fine-tune object detection and fine-grained classification models to identify and localize specific objects, equipment, text, and symbols in publicly available imagery
• Establish evaluation sets and metrics (precision/recall, mAP, IoU) to measure performance and target improvements against real-world cases
• Manage the annotation workflow — defining labeling schemes and directing external data-labeling vendors and tools
• Research and apply current computer vision methods to problems that don't have an off-the-shelf solution
• Improve model robustness to messy, real-world web imagery (varied resolution, compression, watermarks, screenshots)
• Contribute to the broader machine learning stack, including the NLP and large language models behind summarization, classification, and scoring
• Collaborate closely with the engineering team to integrate models into the platform and keep computer vision and platform ML aligned
• Monitor model performance in production and iterate through active-learning and retraining loops

QUALIFICATIONS

• 5+ years in machine learning, with deep hands-on computer vision experience
• Proven track record training, fine-tuning, and evaluating object detection and image classification models
• Strong data-engineering skills — building web scrapers and crawlers, and assembling large-scale datasets from raw, unstructured web sources
• Experience wrangling large image datasets at scale: cleaning, deduplication, filtering, and storage
• Strong Python and deep-learning framework experience (PyTorch and/or TensorFlow)
• Strong software-engineering fundamentals — writes production-quality Python, uses version control, and ships models as maintainable, deployable code (not one-off notebooks)
• Working knowledge of modern detection and vision architectures (YOLO family, DETR variants, vision transformers) and fine-grained classification techniques
• Experience defining annotation schemes and managing labeling workflows or vendors
• Rigorous approach to model evaluation and error analysis (precision/recall, mAP, IoU)
• Experience deploying and serving models in a cloud environment (AWS)
• General machine learning breadth beyond computer vision, including familiarity with NLP / transformer / large language models
• Comfortable working across the stack and moving between problem areas as priorities shift — computer vision is the focus, but this role does not stay in one lane
• Ability to take an ambiguous, unsolved problem and drive it to a working, measurable solution
• Strong written and verbal communication — able to explain approaches and tradeoffs to both engineers and non-technical stakeholders

• Experience with imagery sourced from the open web or social media, as opposed to clean sensor or studio data
• Large-scale data pipeline and workflow orchestration experience (e.g., Airflow, or similar)
• Text and symbol recognition in images (OCR, logo/insignia detection)
• Familiarity with foundation-model-assisted labeling (e.g., Grounding DINO, Segment Anything) and active learning / weak supervision
• Experience with dataset-curation and evaluation tooling (e.g., FiftyOne) and labeling platforms (e.g., Roboflow, Scale AI, Labelbox)
• MLOps experience: experiment tracking, model versioning, reproducible training pipelines
• Background in OSINT, defense, national security, or trust & safety
• Multimodal and vector-search experience