Lead AI and ML Engineer

Christine Straub

ML/AI Engineer — GPU Inference & Model Serving — LLMOps — MLOps

949-***-**** # ****************@*****.*** ï straubchristine § christinestraub christinemstraub San Clemente, CA

Summary

Lead AI/ML Engineer with 8+ years building production AI systems across defense, healthcare, finance, and other enterprise domains. Hands-on depth in GPU inference optimization and model serving—self-hosted LLM/VLM deployment with vLLM (continuous batching, KV-cache tuning), quantization (AWQ/GGUF, FP16/BF16), and ONNX/TensorRT acceleration, backed by rigorous throughput/latency benchmarking on A100/H100 GPUs. Also specializes in LLM/VLM productization, applied agentic workflows, and MLOps/LLMOps infrastructure, including advanced RAG pipelines and multi-agent orchestration systems. Education

University of California, Berkeley Berkeley, CA

Bachelor of Arts in Computer Science December 2017 University of California, Berkeley Berkeley, CA

Bachelor of Arts in Cognitive Science December 2017 Work Experience

Lead AI/Machine Learning Engineer July 2025 – Present Medici Land Governance Washington, DC

Develops blockchain-based land administration and titling systems to modernize property ownership records.

• Gemini Document Intelligence Pipeline: Designed and productionized an end-to-end Gemini (Vertex AI) document processing pipeline for legal land records—deeds, liens, mortgages, and court dockets—combining batch and realtime inference with tiered escalation (Flash/Lite Pro) to maximize throughput while controlling cost.

• Open-Source VLM Experimentation (Qwen): Fine-tuned and evaluated open-source vision-language models

(frontier-Qwen3-model VL via cascade PEFT/to LoRA) identify against the most proprietary cost-effective APIs; path integrated per document self-hosted difficulty. inference into a hybrid OCR VLM

• Self-Hosted GPU Inference & Serving (vLLM, A100/H100): Deployed Qwen3-VL 32B on A100/H100 GPUs with vLLM, tuning continuous batching, KV-cache utilization, and max-concurrency settings; ran systematic concurrency/latency/throughput benchmarks to size production capacity and validate cost-per-page against proprietary APIs, with schema-enforced structured outputs (Instructor + Pydantic) in a two-stage extraction design.

• Cost-Optimized Multi-Model Routing: Built a dynamic routing layer across Gemini, Claude, and open-source VLMs that routes clean pages to cheaper models and escalates only on handwriting/degraded scans—cutting inference costs by

70% while holding 95%+ field extraction accuracy on complex legal documents.

• Core Technologies: Gemini (Vertex AI Batch/Realtime), Qwen Models, Anthropic Models, GCP (Cloud Run, BigQuery, Vertex AI, Document AI, GKE), PaddleOCR, Kedro, vLLM, Kafka, Elasticsearch, MLflow. Lead AI/Machine Learning Engineer January 2026 – June 2026 Bespokelabs AI Mountain View, CA

Applied AI research lab pioneering data and RL environment curation for training and evaluating agents.

• RL Environment Design & Task Authoring: Lead creation of high-quality RL training and evaluation environments for multi-turn tool-calling agents, authoring task specifications, verifier functions, and reward signals, developing repeatable recipes that scaled validated environment production while maintaining rigorous quality bars.

• Agent Failure Analysis & Benchmark Development: Analyzed agent rollouts across diverse task types to diagnose reward hacking and surface failure modes, translating insights into externally released benchmark suites packaged for the company’s evaluation platform and external-facing dashboards.

• Core Technologies: Python, PyTorch, JAX, Hugging Face (Transformers, Datasets, TRL), vLLM, Verifiers, Bespoke Curator, Docker, GCP.

Senior Machine Learning Engineer 2022-2025

RIOS Intelligent Machines Palo Alto, CA

Builds computer vision and ML infrastructure for robotic process automation in manufacturing and industrial environments.

• End-to-End MLOps Pipeline: Architected Metaflow workflow reducing deployment time by 70%, enabling 60 FPS real-time inference via Kubernetes auto-scaling for industrial video anomaly detection processing 10M+ daily images.

• Active Learning Loop: Integrated FiftyOne and Encord to automate data selection, reducing manual annotation time by 60% via uncertainty sampling and bidirectional annotation-dataset pipeline.

• GPU Inference Optimization (TensorRT/ONNX): Designed custom quantized computer vision operators (YOLO v8/v9), achieving 5x inference speedup on GPU edge devices (Jetson/Orin) through ONNX graph optimization and TensorRT engine compilation and quantization.

• Core Technologies: YOLO v8/v9, Metaflow, FiftyOne, Encord, PyTorch, Kubernetes, Docker, Weights & Biases, ONNX, TensorRT.

Senior AI/Machine Learning Engineer May 2023 – April 2025 Unstructured IO San Francisco, CA

Develops open-source and API-based tools that transform unstructured documents into structured data.

• Multi-Agent AI System: Architected production Pydantic AI + MCP (Model Context Protocol) orchestration system, creating custom MCP servers interfacing with unstructured APIs for intelligent document processing workflows.

• VLM Benchmarking & Integration: Evaluated 10+ VLM providers (Claude 3.5/3.7, GPT-4o, Gemini 1.5/2.0), engineered prompt optimization techniques dramatically improving table structure recognition and image text extraction accuracy.

• Layout Detection Fine-Tuning: Fine-tuned layout models (YOLOX, YOLO-NAS, Detectron2) on 11,000+ technical PDFs, achieving 10% accuracy increase and 13% reduction in missing text detection.

• OCR Pipeline Optimization: Led OCR enhancement across Tesseract/PaddleOCR, implementing preprocessing

(scaling, contrast, PSM optimization) improving accuracy by 15% and reducing missing text by 10%.

• Production Impact: Resolved 300+ bugs, reviewed 500+ PRs, implemented memory optimizations (chunked PDF processing), intelligent encoding detection for enterprise-scale processing.

• Core Technologies: YOLOX/YOLO-NAS/Detectron2, Tesseract/PaddleOCR, OpenCV, LangChain, LlamaIndex, Claude/GPT/Gemini APIs, Pydantic AI.

Lead Software Engineer – AI/ML (DoD) Jan 2021 – Feb 2024 Sapient Logic San Diego, CA

AI-enabled intelligence systems for the Department of Defense across multiple classified programs.

• Mobile Intelligence Translation App: Architected mobile OCR platform enabling real-time document processing in air-gapped field environments using Google ML Kit + Tesseract4Android for French/Arabic/Chinese materials during sensitive site exploitation operations.

• AI-Powered Intelligence Management Platform: Led development of microservices-based intelligence requirements system, engineering semi-automated validation workflows and ML-powered classification that streamlined intelligence processing across security classifications, integrating with GCGS-J, ICSF, and Tactical Awareness Kit.

• Cybersecurity Threat Intelligence App: Developed BERT/SBERT-based semantic similarity engine automating MITRE ATT&CK framework mapping to defense mechanisms, reducing manual security gap analysis time by 94% through multi-vector NLP pipeline (Word2Vec, GloVe, Transformers).

• Technical Leadership: Directed cross-functional teams through full SDLC for classified systems, created architecture documentation for defense applications in network-denied tactical environments, and established test management frameworks for multiple DoD programs.

• Core Technologies: Google ML Kit, Tesseract OCR, BERT/SBERT, PyTorch, Word2Vec/GloVe, Microservices, Docker, React, MITRE ATT&CK, Air-gapped systems, Cross-domain solutions. Lead Software Engineer – AI/ML Infrastructure April 2021 – Feb 2023 Sapient Logic San Diego, CA

HIPAA-compliant Electronic Health Record system for hospital patient management.

• HIPAA-Compliant Cloud Architecture: Architected complete EHR solution on AWS (EC2, RDS, VPC, ECS, ECR) with comprehensive security controls including KMS encryption, MFA (Vonage/SendGrid), role-based access, audit logging via CloudWatch.

• Patient Data Management: Engineered comprehensive platform tracking medical histories, demographics, documents from registration through discharge, with real-time ER analytics dashboards for patient flow optimization.

• Healthcare Interoperability: Integrated CollaborateMD, eClaimStatus, epowerdoc using HL7/FHIR protocols for seamless patient management and claims processing.

• Technical Leadership & Team Management: Led engineering team implementing physician-requested improvements, conducted technical interviews and hired full-stack, DevOps, and QA engineers, created comprehensive product requirement documentation, and established GitHub Actions CI/CD pipeline with DevOps collaboration.

• Core Technologies: AWS (EC2, RDS, VPC, ECS, ECR, CloudWatch), Django, Vue.js, React, PostgreSQL, HL7/FHIR, Docker, GitHub Actions.

AI Software Architect June 2022 – June 2023

Speechlab AI San Francisco, CA

AI-powered audio/video localization platform enabling speech-to-speech translation and synthetic dubbing.

• Cloud-Native Localization Architecture: Designed end-to-end AWS serverless infrastructure (Lambda, EventBridge, ECS, App Runner, S3) for event-driven media processing, reducing dubbing costs by 80% and accelerating timelines from weeks to hours.

• ML API Gateway: Built scalable API layer integrating speech recognition, machine translation, and text-to-speech ML models with performance optimization and model monitoring for reliable multilingual workflows.

• Core Technologies: AWS (Lambda, EventBridge, S3, ECS, App Runner, Cognito, CloudWatch, SES), Node.js, Express.js, React, TypeScript, MongoDB.

Machine Learning Engineer May 2021 – May 2022

Collegis Education Chicago, IL

Data analytics solutions optimizing enrollment growth and student success for higher education institutions.

• Event-Driven ETL Pipeline: Architected near-real-time pipelines using Google Cloud Functions/Cloud Run ingesting student interaction data from Phoneburner, Five9, LMS Canvas, reducing data latency by 85%.

• Speech Analytics: Engineered Call Center Intelligence System using Cloud Speech-to-Text and Natural Language APIs for automated transcription and sentiment analysis.

• Core Technologies: GCP (Cloud Functions, Run, BigQuery, Speech-to-Text, Natural Language API), DBT, Fivetran, Python, ThoughtSpot.

Software Engineer September 2017 – April 2021

Moody's Analytics Silicon Valley, CA

Catastrophe risk modeling and location intelligence platforms for insurance and financial services clients.

• Software Engineering: Contributed to RMS(one), RiskLink, and RiskBrowser platforms, building client-facing features for catastrophe risk assessment and portfolio analysis enabling insurers to evaluate multi-billion dollar portfolios.

• Core Technologies: Java, TypeScript, Python, Geospatial visualization, Data pipelines, Risk modeling. Technical Skills

GPU Inference & Model Serving: vLLM (Continuous Batching, PagedAttention, KV-Cache Tuning), TensorRT, ONNX Runtime, Ray Serve, Quantization (AWQ, GGUF, INT8, FP16/BF16 Mixed Precision), Throughput/Latency/Concurrency Benchmarking (A100/H100), Self-Hosted LLM/VLM Deployment (RunPod, GKE) Agentic AI & Generative AI (GenAI): AI Agents & Multi-Agent Systems (MAS), Agentic Workflows, Retrieval-Augmented Generation (RAG: GraphRAG/HyDE), Large Language Models (LLMs), Vision-Language Models

(VLMs), ReAct, Chain-of-Thought (CoT), Prompt Engineering & Optimization (DSPy), Model Context Protocol (MCP), Autonomous Agents

ML/AI/NLP Expertise: Computer Vision (Object Detection/Segmentation), Natural Language Processing (NLP), Transformer Architectures (BERT/ViT), Reinforcement Learning (RL), Model Fine-tuning (LoRA/QLoRA), Quantization

(GGUF/AWQ), OCR, Edge AI, Distributed Training

Agentic Frameworks & LLMs: LangGraph, CrewAI, Microsoft AutoGen, Pydantic AI, LangChain, LlamaIndex, Claude

(Sonnet 4.6/Opus 4.x/Fable 5), GPT-4o/o3/GPT-5.x, Gemini 2.5/3.x (Flash/Pro), Qwen3/Qwen3-VL, Llama 3.x, Ollama, vLLM

LLM Evaluation, Benchmarking & Observability: Ragas (RAG Assessment), LangSmith, Arize Phoenix (LLM Observability), DeepEval, RL Environment Design & Agent Benchmarking, Human-in-the-loop (HITL) Evaluation, Hallucination Detection, Bias & Safety Testing, Red-Teaming Deep Learning & MLOps: PyTorch, TensorFlow, Keras, Hugging Face (Transformers/TRL), ONNX, TensorRT, Metaflow, MLflow (Model Versioning/Registry), Weights & Biases (Experiment Tracking), Ray Serve, Model Monitoring, Inference Cost Optimization, CI/CD (GitHub Actions) Computer Vision: YOLO (v8/v9/NAS), Detectron2, Tesseract, PaddleOCR, OpenCV, FiftyOne, Encord, Google ML Kit, SAM (Segment Anything Model)

Cloud & Infrastructure: AWS (Lambda, SageMaker, Bedrock, ECS, EventBridge), GCP (Vertex AI, Cloud Run, BigQuery, Document AI), Kubernetes (K8s), Docker, Terraform Fullstack & Databases: Python, SQL, Node.js, TypeScript, Rust, Microservices & REST APIs, PostgreSQL, MongoDB, Pinecone/Weaviate/ChromaDB (Vector Search), Elasticsearch, Kafka 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