San Clemente, California, United States
Lead Machine Learning/AI Engineer with 8+ years building production AI systems across defense, healthcare, and enterprise software. Working on Reinforcement Learning environment curation and agent training, GRPO-based agent training pipelines, difficulty-calibrated Reinforcement Learning environment design, and data curation at scale. Deep experience in LLMOps, MLOps, Evaluation Pipelines, Multi-agent orchestration, and VLM pipelines across HIPAA and DoD environments. Core Expertise: - RL Environment Curation - Agent Training & Evaluation - Data Curation at Scale - Web-Scale LLMOps - Document Intelligence - LLM Orchestration - Evaluation Pipelines - RAG ๐ Portfolio: https://christinemstraub.com ๐ GitHub: https://github.com/christinemstraub ๐ Notion: https://christinestraub.notion.site/ml
- Design and build RL environments and reward functions that produce clean, learnable signals for frontier models on ML research and engineering tasks. - Build deep expertise across the frontier of ML research, training, and inference infrastructure. - Collaborate with others to brainstorm and create new ideas and tools to improve the environment-building process
โข ๐๐๐ถ๐น๐ ๐ฝ๐ฟ๐ผ๐ฑ๐๐ฐ๐๐ถ๐ผ๐ป ๐ ๐๐ข๐ฝ๐ ๐ถ๐ป๐ณ๐ฟ๐ฎ๐๐๐ฟ๐๐ฐ๐๐๐ฟ๐ฒ using ๐ ๐ฒ๐๐ฎ๐ณ๐น๐ผ๐ workflows and ๐๐๐ฏ๐ฒ๐ฟ๐ป๐ฒ๐๐ฒ๐ for ๐ฌ๐ข๐๐ข-๐ฏ๐ฎ๐๐ฒ๐ฑ ๐ฐ๐ผ๐บ๐ฝ๐๐๐ฒ๐ฟ ๐๐ถ๐๐ถ๐ผ๐ป ๐ฝ๐ฟ๐ผ๐ฐ๐ฒ๐๐๐ถ๐ป๐ด. ย ย โข Refactored standalone scripts into structured ๐ฝ๐ฎ๐ฟ๐ฎ๐น๐น๐ฒ๐น ๐ฝ๐ฟ๐ผ๐ฐ๐ฒ๐๐๐ถ๐ป๐ด ๐ฎ๐ฟ๐ฐ๐ต๐ถ๐๐ฒ๐ฐ๐๐๐ฟ๐ฒ with FrameServerHandler. Reduced processing latency by 75% with real-time performance optimization. ย ย โข Developed "Thoth v0" ๐ฑ๐ฎ๐๐ฎ ๐๐ฐ๐ถ๐ฒ๐ป๐ฐ๐ฒ ๐ฝ๐น๐ฎ๐๐ณ๐ผ๐ฟ๐บ with bidirectional ๐๐ถ๐ณ๐๐๐ข๐ป๐ฒ-๐๐ป๐ฐ๐ผ๐ฟ๐ฑ ๐ถ๐ป๐๐ฒ๐ด๐ฟ๐ฎ๐๐ถ๐ผ๐ป for ๐บ๐ฎ๐ฐ๐ต๐ถ๐ป๐ฒ ๐น๐ฒ๐ฎ๐ฟ๐ป๐ถ๐ป๐ด ๐๐ผ๐ฟ๐ธ๐ณ๐น๐ผ๐๐. Created 5-step automated ๐ฎ๐ป๐ป๐ผ๐๐ฎ๐๐ถ๐ผ๐ป ๐ฝ๐ถ๐ฝ๐ฒ๐น๐ถ๐ป๐ฒ from data sampling to model retraining. Accelerated Machine Learning cycles by 65% using ๐๐ช๐ฆ ๐ฆ๐ฏ and ๐ ๐ผ๐ป๐ด๐ผ๐๐ backend. ย ย โข ๐๐ป๐ด๐ถ๐ป๐ฒ๐ฒ๐ฟ๐ฒ๐ฑ ๐ณ๐ฎ๐ฐ๐๐ผ๐ฟ๐ ๐ฝ๐ฎ๐๐๐ฒ๐ฟ๐ป ๐ฑ๐ฎ๐๐ฎ ๐ฎ๐ฟ๐ฐ๐ต๐ถ๐๐ฒ๐ฐ๐๐๐ฟ๐ฒ with ๐ฑ๐๐ป๐ฎ๐บ๐ถ๐ฐ ๐ฏ๐ฎ๐๐ฐ๐ต๐ถ๐ป๐ด for sequential video processing. Implemented sequence_id-based organization with timestamp ordering for ๐๐ถ๐บ๐ฒ-๐๐ฒ๐ฟ๐ถ๐ฒ๐ ๐ฑ๐ฒ๐ฒ๐ฝ ๐น๐ฒ๐ฎ๐ฟ๐ป๐ถ๐ป๐ด models. ย ย โข ๐ข๐ฝ๐๐ถ๐บ๐ถ๐๐ฒ๐ฑ ๐ฑ๐ฎ๐๐ฎ ๐ฝ๐ถ๐ฝ๐ฒ๐น๐ถ๐ป๐ฒ performance from days to hours for robotics applications. ย ย โข ๐๐ฟ๐ฒ๐ฎ๐๐ฒ๐ฑ ๐ฐ๐๐๐๐ผ๐บ ๐๐/๐ ๐ ๐๐ผ๐ผ๐น๐ถ๐ป๐ด including "Thoth Frames Retriever" FiftyOne plugin with multi-camera support and ๐๐ฃ๐จ-๐ฎ๐ฐ๐ฐ๐ฒ๐น๐ฒ๐ฟ๐ฎ๐๐ฒ๐ฑ ๐ฝ๐ฟ๐ผ๐ฐ๐ฒ๐๐๐ถ๐ป๐ด. ย ย โข Built 5 core computer vision operators using VOXEL51 Plugins for model deployment and segmentation training. Improved team productivity by ~50% through cross-functional collaboration. ย ย โข ๐๐ฟ๐ฐ๐ต๐ถ๐๐ฒ๐ฐ๐๐ฒ๐ฑ ๐ฑ๐ถ๐๐๐ฟ๐ถ๐ฏ๐๐๐ฒ๐ฑ ๐๐๐๐๐ฒ๐บ๐ with ๐๐ผ๐ฐ๐ธ๐ฒ๐ฟ containerization and ๐๐ถ๐ฑ๐ฒ๐ผ ๐๐๐ถ๐๐ฐ๐ต๐ถ๐ป๐ด capabilities for ๐ฒ๐ฑ๐ด๐ฒ ๐๐ ๐ฎ๐ฝ๐ฝ๐น๐ถ๐ฐ๐ฎ๐๐ถ๐ผ๐ป๐. Implemented ๐ต๐ถ๐ด๐ต-๐ฝ๐ฒ๐ฟ๐ณ๐ผ๐ฟ๐บ๐ฎ๐ป๐ฐ๐ฒ ๐ฐ๐ผ๐บ๐ฝ๐๐๐ถ๐ป๐ด with configurable parameters and ๐ฐ๐ผ๐บ๐ฝ๐ฟ๐ฒ๐ต๐ฒ๐ป๐๐ถ๐๐ฒ ๐น๐ผ๐ด๐ด๐ถ๐ป๐ด.
๐๐ฒ๐ฑ ๐ฑ๐ฒ๐๐ฒ๐น๐ผ๐ฝ๐บ๐ฒ๐ป๐ ๐ผ๐ณ ๐๐๐ผ ๐ฐ๐ฟ๐ถ๐๐ถ๐ฐ๐ฎ๐น ๐๐ฒ๐ฝ๐ฎ๐ฟ๐๐บ๐ฒ๐ป๐ ๐ผ๐ณ ๐๐ฒ๐ณ๐ฒ๐ป๐๐ฒ ๐๐ ๐ฝ๐ฟ๐ผ๐๐ผ๐๐๐ฝ๐ฒ๐ (๐ฃ๐ข๐๐๐ฅ & ๐๐ข๐ ๐๐ง), ๐ฑ๐ฒ๐น๐ถ๐๐ฒ๐ฟ๐ถ๐ป๐ด ๐ฝ๐ฟ๐ผ๐ฑ๐๐ฐ๐๐ถ๐ผ๐ป-๐ฟ๐ฒ๐ฎ๐ฑ๐ ๐บ๐ฎ๐ฐ๐ต๐ถ๐ป๐ฒ ๐น๐ฒ๐ฎ๐ฟ๐ป๐ถ๐ป๐ด ๐๐๐๐๐ฒ๐บ๐ ๐๐ต๐ฎ๐ ๐ฒ๐ป๐ต๐ฎ๐ป๐ฐ๐ฒ๐ฑ ๐บ๐ถ๐น๐ถ๐๐ฎ๐ฟ๐ ๐ถ๐ป๐๐ฒ๐น๐น๐ถ๐ด๐ฒ๐ป๐ฐ๐ฒ ๐ผ๐ฝ๐ฒ๐ฟ๐ฎ๐๐ถ๐ผ๐ป๐ ๐ณ๐ผ๐ฟ ๐ญ๐ฌ,๐ฌ๐ฌ๐ฌ+ ๐ฎ๐ฐ๐๐ถ๐๐ฒ ๐๐๐ฒ๐ฟ๐ ๐ฎ๐ฐ๐ฟ๐ผ๐๐ ๐๐ฎ๐ฐ๐๐ถ๐ฐ๐ฎ๐น ๐ฒ๐ป๐๐ถ๐ฟ๐ผ๐ป๐บ๐ฒ๐ป๐๐. ๐ฃ๐ข๐๐๐ฅ - ๐ ๐ผ๐ฏ๐ถ๐น๐ฒ ๐๐ผ๐ฐ๐๐บ๐ฒ๐ป๐ ๐๐ป๐๐ฒ๐น๐น๐ถ๐ด๐ฒ๐ป๐ฐ๐ฒ ๐ฆ๐๐๐๐ฒ๐บ ๐๐ฟ๐ฐ๐ต๐ถ๐๐ฒ๐ฐ๐๐ฒ๐ฑ ๐ฒ๐ป๐ฑ-๐๐ผ-๐ฒ๐ป๐ฑ ๐บ๐ผ๐ฏ๐ถ๐น๐ฒ ๐ข๐๐ฅ ๐ฝ๐ถ๐ฝ๐ฒ๐น๐ถ๐ป๐ฒ processing 3+ languages (Arabic, Chinese, French) with 95%+ accuracy for tactical document capture in ๐ป๐ฒ๐๐๐ผ๐ฟ๐ธ-๐ฑ๐ฒ๐ป๐ถ๐ฒ๐ฑ ๐ฒ๐ป๐๐ถ๐ฟ๐ผ๐ป๐บ๐ฒ๐ป๐๐. ๐๐บ๐ฝ๐น๐ฒ๐บ๐ฒ๐ป๐๐ฒ๐ฑ ๐ฟ๐ฒ๐ฎ๐น-๐๐ถ๐บ๐ฒ ๐ ๐ ๐๐ฟ๐ฎ๐ป๐๐น๐ฎ๐๐ถ๐ผ๐ป ๐๐๐๐๐ฒ๐บ using ๐๐ผ๐ผ๐ด๐น๐ฒ ๐ ๐ ๐๐ถ๐ and ๐ง๐ฒ๐๐๐ฒ๐ฟ๐ฎ๐ฐ๐๐ฐ๐๐ป๐ฑ๐ฟ๐ผ๐ถ๐ฑ, reducing ๐ณ๐ผ๐ฟ๐ฒ๐ถ๐ด๐ป ๐น๐ฎ๐ป๐ด๐๐ฎ๐ด๐ฒ ๐ฑ๐ผ๐ฐ๐๐บ๐ฒ๐ป๐ ๐ฝ๐ฟ๐ผ๐ฐ๐ฒ๐๐๐ถ๐ป๐ด time from hours to seconds. ๐๐ฒ๐๐ถ๐ด๐ป๐ฒ๐ฑ ๐๐ฒ๐ฐ๐๐ฟ๐ฒ ๐ฐ๐ฟ๐ผ๐๐-๐ฑ๐ผ๐บ๐ฎ๐ถ๐ป ๐ฑ๐ฎ๐๐ฎ ๐ณ๐น๐ผ๐๐ enabling one-way transfer of classified intelligence to ๐๐ผ๐บ๐บ๐ผ๐ป ๐ข๐ฝ๐ฒ๐ฟ๐ฎ๐๐ถ๐ผ๐ป๐ฎ๐น ๐ฃ๐ถ๐ฐ๐๐๐ฟ๐ฒ (๐๐ข๐ฃ) ๐๐๐๐๐ฒ๐บ๐ while maintaining ๐๐ข๐ ๐๐ฒ๐ฐ๐๐ฟ๐ถ๐๐ ๐ฐ๐ผ๐บ๐ฝ๐น๐ถ๐ฎ๐ป๐ฐ๐ฒ. ๐ข๐ฝ๐๐ถ๐บ๐ถ๐๐ฒ๐ฑ ๐บ๐ผ๐ฏ๐ถ๐น๐ฒ ๐ ๐ ๐ฝ๐ฒ๐ฟ๐ณ๐ผ๐ฟ๐บ๐ฎ๐ป๐ฐ๐ฒ for field-deployed Android devices, achieving sub-2-second ๐ข๐๐ฅ ๐ฝ๐ฟ๐ผ๐ฐ๐ฒ๐๐๐ถ๐ป๐ด for combat-ready intelligence gathering. ๐๐ข๐ ๐๐ง - ๐๐ป๐๐ฒ๐น๐น๐ถ๐ด๐ฒ๐ป๐ฐ๐ฒ ๐ฅ๐ฒ๐พ๐๐ถ๐ฟ๐ฒ๐บ๐ฒ๐ป๐๐ ๐ ๐ฎ๐ป๐ฎ๐ด๐ฒ๐บ๐ฒ๐ป๐ ๐ฃ๐น๐ฎ๐๐ณ๐ผ๐ฟ๐บ ๐๐๐ถ๐น๐ ๐บ๐ถ๐ฐ๐ฟ๐ผ๐๐ฒ๐ฟ๐๐ถ๐ฐ๐ฒ๐-๐ฏ๐ฎ๐๐ฒ๐ฑ ๐ถ๐ป๐๐ฒ๐น๐น๐ถ๐ด๐ฒ๐ป๐ฐ๐ฒ ๐ฝ๐น๐ฎ๐๐ณ๐ผ๐ฟ๐บ serving 500+ analysts across multiple security classifications with 99.9% uptime SLA. ๐๐ฒ๐๐ฒ๐น๐ผ๐ฝ๐ฒ๐ฑ ๐๐ฒ๐บ๐ถ-๐ฎ๐๐๐ผ๐บ๐ฎ๐๐ฒ๐ฑ ๐ฐ๐ผ๐น๐น๐ฒ๐ฐ๐๐ถ๐ผ๐ป ๐ฎ๐น๐ด๐ผ๐ฟ๐ถ๐๐ต๐บ๐ reducing intelligence validation cycles by 70% through ML-powered requirement classification and routing.