Post Job Free
Sign in

Lead AI and ML Engineer

Location:
San Clemente, CA
Posted:
November 20, 2026

Contact this candidate

Resume:

Christine Straub

Lead Machine Learning Engineer AI Engineer Document AI, LLM Inference & Agent Evaluation

949-***-**** ****************@*****.*** Los Angeles, CA (Pacific Time) linkedin.com/in/straubchristine github.com/christinestraub christinemstraub.com Summary

Lead Machine Learning Engineer and AI Engineer with 9+ years shipping production document AI, LLM evaluation and MLOps in defense, healthcare and legal tech, leading teams of 4–20 engineers. Served Qwen3-VL 32B on vLLM

(A100/H100) as the first-line vision-language model (VLM), cutting per-page extraction cost about 25x at 94.3% token-F1. Led 15–20 engineers building reinforcement learning (RL) environments that frontier AI labs use to evaluate large language model (LLM) coding agents.

Education

University of California, Berkeley Berkeley, CA

Bachelor of Arts in Computer Science Dec 2017

University of California, Berkeley Berkeley, CA

Bachelor of Arts in Cognitive Science Dec 2017

Skills

Generative AI & LLMs: LLMs, VLMs, Qwen3-VL, Gemini, Claude, GPT-4o, Llama; retrieval-augmented generation

(RAG), AI agents, Model Context Protocol (MCP), LangChain, LangGraph, prompt engineering, embeddings, vector search (Azure AI Search)

Model Serving & Inference: vLLM, SGLang, TensorRT-LLM, ONNX Runtime; continuous batching, KV-cache tuning, quantization (AWQ, GGUF, FP16/BF16); NVIDIA A100/H100 GPUs LLM Evaluation & RL: LLM-as-a-judge, RAGAS, agent evaluation, red-teaming, prompt injection detection; RL environments, verifiers, reward signals, Reinforcement Learning from Human Feedback (RLHF), Proximal Policy Optimization (PPO), Group Relative Policy Optimization (GRPO), Direct Preference Optimization (DPO) Document AI & Computer Vision: Optical character recognition (OCR), PaddleOCR, Tesseract, YOLOX, YOLO-NAS, Detectron2, layout detection, intelligent document processing (IDP) ML & MLOps: PyTorch, Hugging Face (Transformers), fine-tuning (Low-Rank Adaptation (LoRA)), Bidirectional Encoder Representations from Transformers (BERT), DeepSpeed, Megatron-LM, XGBoost, MLflow, Metaflow, Extract, Transform, Load (ETL), Kubernetes, Docker, Continuous Integration / Continuous Deployment (CI/CD) Cloud & Languages: Google Cloud (GCP: Vertex AI, Google Kubernetes Engine (GKE), BigQuery), Amazon Web Services (AWS), Azure AI Search; Python, SQL, dbt, TypeScript, Java, FastAPI Healthcare & NLP: Health Insurance Portability and Accountability Act (HIPAA), Health Level Seven (HL7), Fast Healthcare Interoperability Resources (FHIR), electronic health records (EHR), natural language processing (NLP) Work Experience

Wider Security Aug 2026 – Present

Senior Machine Learning Engineer, AI Security · Contract Remote AEGIS: a patent-pending runtime security layer that supervises AI agents and blocks harmful actions.

• AI Agent Threat Detection: Build prompt injection and social engineering detectors pairing static pre-checks with LLM-as-a-judge scoring; in shadow mode they cut false positives by more than half.

• Evaluation & Training Harness: Designed and built the offline evaluation harness in Python, with go-live benchmarks of 1,000+ red-team agent sessions; now benchmarking the latest models (Gemma 4, Llama).

• Agent Security Hardening: Building signed, fail-closed content provenance that separates trusted operator instructions from untrusted tool, file and web content, plus isolated environments for adversarial testing. Medici Land Governance Jul 2025 – Nov 2026

Lead Machine Learning Engineer, Document AI · Contract Remote, United States

• Difficulty-Aware Routing & Serving: Made self-hosted Qwen3-VL 32B on vLLM (A100/H100, GKE) the first-line model, with Claude and Gemini behind it, cutting per-page extraction cost about 25x at 94.3% token-F1.

• Historical Records: Cut manual review by 40% with a deep-learning pipeline for millions of 1850s-era court dockets.

• LLMOps & Leadership: Technical lead for 4 engineers; built the evaluation service gating releases against a measured noise floor; cut never-processed-document detection from about 50 minutes to under a minute. Bespoke Labs Jan 2026 – Aug 2026

Lead Machine Learning Engineer, RL Environments & Agent Evaluation · Freelance Remote, United States

• RL Environment Design: Led 15–20 engineers building two RL environments that frontier AI labs use to train and evaluate LLM coding agents; authored task specifications, verifier functions and reward signals.

• Agent Evaluation: Led ML engineering for a platform with 100+ RL tasks and 10,000+ sandboxed rollouts across 5 LLM providers; XGBoost pass-rate prediction rejected 10%+ of tasks, saving 40+ hours/week.

• Benchmark Hardening: As majority contributor to the ML platform benchmark, built CI promotion gates, anti-cheating checks and Dockerfile security lockdown.

RIOS Intelligent Machines Jun 2022 – May 2025

Senior Machine Learning Engineer, ML Data Platform & Computer Vision · Contract Palo Alto, CA

• Data Labeling: Built a FiftyOne Encord labeling plugin with active learning that cut annotation handoff from hours to minutes for 5,000+ video sequences, plus Metaflow ETL on Kubernetes (Docker).

• Computer Vision Data Formats: Shipped 10 dataset formats for 3D vision, object detection, segmentation and OCR research, cutting new-format data prep from about 1 day to minutes. Unstructured May 2023 – Apr 2025

Senior Machine Learning Engineer, Document AI · Full-time San Francisco Bay Area

• RAG Ingestion Quality: Fine-tuned YOLOX, YOLO-NAS and Detectron2 layout models on 11,000+ PDFs for RAG ingestion, raising accuracy 10%; cut hi_res peak memory from 12.9 GB+ to 848 MB.

• OCR & Reading Order: Led Tesseract and PaddleOCR improvements (scaling, contrast, PSM); implemented XY-cut reading order; raised OCR accuracy by 15% and cut missing text by 10%.

• Agents & Open Source: Built Pydantic AI and MCP agent orchestration; led evaluation of 10+ VLM providers; one of the open-source maintainers (15K+ stars), resolving 300+ bugs, reviewing 1,000+ PRs. Sapient Logic Jan 2021 – Feb 2024

Lead Software Engineer, AI/ML (Defense & Healthcare) · Full-time San Diego, CA

• Defense AI Leadership: Led a team of 5 engineers taking 2 Department of Defense (DoD) AI prototypes to production-ready machine learning systems for 10,000+ military intelligence users.

• Cybersecurity NLP: Cut manual security-gap analysis time by 94% with a BERT/SBERT semantic similarity engine

(PyTorch, TensorFlow) mapping cyber defenses to MITRE ATT&CK at 92% precision.

• HIPAA EHR: Architected and led an electronic health record (EHR) for 10,000+ users with 3 HL7/FHIR integrations. Speechlab AI Jun 2022 – Jun 2023

Senior Software Architect, AI Speech Platform · Contract Remote, United States

• Platform: Cut AI dubbing cost up to 80% and turnaround from weeks to hours with event-driven AWS infrastructure.

• ASR, Translation & TTS: Integrated speech recognition (ASR), machine translation and text-to-speech (TTS) into a multi-accent dubbing pipeline, and designed a GCP data-collection pipeline for training custom TTS models. Collegis Education Apr 2021 – Apr 2022

Senior Machine Learning & Data Engineer · Freelance Remote, United States

• Data Platform: Built serverless ingestion and dbt modeling across 19 partner universities (about 580 SQL models); architected near-real-time GCP pipelines, cutting data latency from hours to minutes.

• NLP Question Taxonomy: Turned chatbot transcripts into a page-level question taxonomy using a dialogue-act classifier, Sentence-Transformers embeddings and k-means clustering (chose 0.872 F1 model over a faster 0.743 one). RMS (Risk Management Solutions), now part of Moody’s Analytics Sep 2017 – Apr 2021 Software Engineer, Catastrophe Risk Platforms · Full-time San Francisco Bay Area

• Built RMS(one), RiskLink and RiskBrowser features (Java, TypeScript, Python) for multi-billion-dollar portfolios. Projects

• Multi-Agent RAG (Duxre, Freelance): Built the orchestrator (Planner, Executor, Presenter) for a real estate assistant powering 15+ revenue-generating workflows, with Azure AI Search vector search and RAGAS gated in CI/CD.

• Coding-Agent Evaluation (AfterQuery, Contract): Author reproducible PyTorch tasks on distributed training

(DeepSpeed, Megatron-LM) and inference engines (vLLM, SGLang, TensorRT-LLM) for evaluating frontier coding agents. Certifications

Deep Learning Specialization (DeepLearning.AI, Stanford) Machine Learning Specialization (DeepLearning.AI, Stanford) MLOps Specialization (DeepLearning.AI) AWS Cloud Practitioner (AWS) IBM Data Science (IBM) Google Data Analytics (Google) Business Intelligence (Google)



Contact this candidate