Skip to content

Christine Straub · Lead Machine Learning & Artificial Intelligence Engineer · Remote / United States

Christine Straub · Lead Machine Learning & Artificial Intelligence Engineer. Models and agents that hold up in the real world.

Deep expertise in reinforcement learning (supervised fine-tuning, direct preference optimization, and group relative policy optimization, plus environment design and verifiers), document intelligence, and multi-agentic workflows.

Open to remote Reinforcement Learning, Post-Training, and Agentic Artificial Intelligence roles

View Case Studies
Download Resume
Contact
  • Reinforcement Learning
  • Post-Training (Supervised Fine-Tuning, Direct Preference Optimization, and Group Relative Policy Optimization)
  • Document Intelligence
  • Multi-Agentic Workflows

Christine Straub.

I'm a Lead Machine Learning & Artificial Intelligence Engineer with 9+ years of experience building and shipping production artificial intelligence systems across defense, healthcare, and enterprise software, including work connected to classified Department of Defense programs and environments regulated by the Health Insurance Portability and Accountability Act.

My core expertise spans the full model lifecycle (pre-training, post-training, and fine-tuning) with deep specialization in reinforcement learning (reinforcement learning from human feedback, supervised fine-tuning, direct preference optimization, and group relative policy optimization, plus environment curation, deterministic verifiers, benchmark development) and multi-agentic workflows that plan, call tools, and hand off work reliably in production.

I'm drawn to artificial intelligence systems where quality depends on more than choosing the right model: strong data curation, rigorous evaluation, reliable infrastructure, thoughtful post-training loops, and product judgment that turns model outputs into real-world value.

Lead Machine Learning & Artificial Intelligence Engineer · Remote / United States

Selected outcomes

years building production artificial intelligence, machine learning, and software systems
9+
PRs reviewed across production systems
500+
production bugs resolved
300+
images/day supported in machine learning pipelines
10M+
deployment-time reduction through machine learning operations automation
70%
infrastructure cost reduction in serverless machine learning and application programming interface architecture
80%
extraction accuracy target in document intelligence workflows
95%+

Production systems, evaluated.

Evaluation infrastructure first, then the agentic, document, and machine learning operations systems it protects. Every case includes measurable outcomes and a technical deep dive.

Reinforcement Learning Environments & Artificial Intelligence Evaluation Infrastructure

Bespokelabs AI · Micro1 · Handshake AI

Designed realistic reinforcement learning environments, task specifications, reward signals, verifiers, hidden tests, and benchmark tasks for artificial intelligence coding agents and model evaluation workflows.

  • Python
  • PyTorch
  • JAX
  • Hugging Face
  • verifiers
  • Docker
  • pytest
  • CI
  • benchmark harnesses

Outcomes

  • Authored realistic multi-step coding-agent repair tasks
  • Built deterministic verification and golden-reference checks
  • Analyzed rollouts for failure modes and reward hacking
  • Created evaluation criteria for reproducibility, correctness, and robustness

Artificial Intelligence Agents, Model Context Protocol Tooling, and Multi-Agent Orchestration

Unstructured IO · Bespokelabs AI · Handshake AI

Built and evaluated agentic systems using Model Context Protocol servers, Pydantic AI, tool-calling workflows, deterministic verification, agent rollouts, and failure-analysis loops.

  • Python
  • TypeScript
  • Model Context Protocol
  • Pydantic AI
  • FastAPI
  • Docker
  • pytest
  • OpenAI
  • Claude
  • Gemini

Outcomes

  • Designed Model Context Protocol-based workflows for tool use and controlled agent execution
  • Evaluated agent failures including wrong tool selection, bad arguments, and reward hacking
  • Created verifier functions, test harnesses, and reproducible benchmark tasks
  • Improved agent reliability through structured evaluation and debugging

Document Intelligence & Multimodal Extraction

Medici Land Governance · Unstructured IO

Built document artificial intelligence systems that turn messy real-world documents — scanned, handwritten, and multi-format — into structured, reviewable data using Gemini Document Intelligence, Qwen vision-language models, optical character recognition pipelines, layout detection, schema-constrained extraction, and human-review workflows.

  • Gemini
  • Vertex AI
  • Qwen3-VL
  • vLLM
  • Instructor
  • Pydantic
  • Optical Character Recognition
  • FastAPI
  • Python
  • GCP

Outcomes

  • Designed hybrid optical character recognition → vision-language model → frontier-model extraction pipelines
  • Used confidence scoring and validation checks for reviewable outputs
  • Supported structured extraction for high-stakes legal and enterprise documents
  • Improved reliability through schema validation and benchmark-driven evaluation
  • Contributed open-source optical character recognition wrappers (PaddleOCR, Tesseract) and parsing pipelines across seven Unstructured repositories

Production Machine Learning Operations, Inference, and Computer Vision Systems

RIOS Intelligent Machines · Sapient Logic · Collegis Education

Built scalable machine learning systems for computer vision, optical character recognition, speech analytics, cloud extract, transform, load, real-time inference, and deployment automation.

  • Python
  • PyTorch
  • YOLO
  • ONNX
  • TensorRT
  • Kubernetes
  • Metaflow
  • GCP
  • AWS
  • Docker

Outcomes

  • Reduced deployment time by 70%
  • Supported real-time 60 frames-per-second machine learning inference workflows
  • Processed 10M+ images/day
  • Reduced manual annotation by 60% through active learning workflows
  • Optimized models for deployment on graphics processing units and edge devices

Timeline.

8 roles across defense, healthcare, finance, robotics, legal document workflows, and enterprise artificial intelligence platforms.

  1. Lead Artificial Intelligence & Machine Learning Engineer · Medici Land Governance

    Document intelligence for complex legal and enterprise documents

    • Built document artificial intelligence for high-stakes legal and enterprise documents
    • Designed hybrid optical character recognition → vision-language model extraction with confidence-scored review
    • Improved reliability through schema validation and benchmark-driven evaluation
    • Gemini
    • Qwen3-VL
    • vLLM
    • Pydantic
    • FastAPI
    • GCP

  2. Lead Artificial Intelligence & Machine Learning Engineer · Bespokelabs AI

    Reinforcement learning environments and evaluation infrastructure for coding agents

    • Designed realistic reinforcement learning environments, task specs, reward signals, and verifiers
    • Analyzed agent rollouts for failure modes and reward hacking
    • Created evaluation criteria for reproducibility, correctness, and robustness
    • Python
    • PyTorch
    • JAX
    • verifiers
    • Docker
    • CI

  3. Senior Machine Learning Engineer · RIOS Intelligent Machines

    Computer vision and robotics for industrial automation

    • Built scalable machine learning systems for computer vision and real-time inference
    • Reduced deployment time by 70% through machine learning operations automation
    • Processed 10M+ images/day and optimized models for deployment on graphics processing units and edge devices
    • PyTorch
    • YOLO
    • TensorRT
    • ONNX
    • Kubernetes
    • Metaflow

  4. Senior Artificial Intelligence & Machine Learning Engineer · Unstructured IO

    Enterprise document processing and agentic workflows

    • Built multimodal extraction pipelines for complex enterprise documents
    • Designed Model Context Protocol-based workflows for tool use and controlled agent execution
    • Created verifier functions, test harnesses, and reproducible benchmark tasks
    • Model Context Protocol
    • Pydantic AI
    • OpenAI
    • Claude
    • FastAPI
    • pytest

  5. Lead Software Engineer / Artificial Intelligence & Machine Learning · Sapient Logic

    Machine learning platform engineering and cloud extract, transform, load

    • Built cloud extract-transform-load pipelines and deployment automation for production machine learning
    • Reduced infrastructure cost by 80% in serverless machine learning and application programming interface architecture
    • Optimized models for deployment on graphics processing units and edge devices
    • Python
    • GCP
    • AWS
    • Docker
    • serverless

  6. Artificial Intelligence Software Architect · Speechlab AI

    Speech analytics and audio intelligence

    • Built scalable machine learning systems for speech analytics
    • Supported real-time inference workflows in production
    • Improved reliability through structured evaluation and debugging
    • Python
    • FastAPI
    • Docker
    • GCP

  7. Machine Learning Engineer · Collegis Education

    Applied machine learning for education platforms

    • Built production machine learning pipelines for cloud extract-transform-load and inference
    • Reduced manual annotation by 60% through active learning workflows
    • Reduced deployment time through machine learning operations automation
    • Python
    • PyTorch
    • Metaflow
    • AWS

  8. Software Engineer · Moody's Analytics

    Financial data platforms and analytics software

    • Built production software systems for financial analytics
    • Reviewed 500+ PRs across production systems
    • Resolved 300+ production bugs
    • Python
    • PostgreSQL
    • Application Programming Interfaces
    • distributed systems

Technical matrix.

The tooling behind the systems above, grouped by how it is used in production.

Reinforcement Learning & Post-Training

  • Reinforcement Learning from Human Feedback
  • Supervised Fine-Tuning
  • Direct Preference Optimization
  • Group Relative Policy Optimization
  • Reinforcement Learning environments
  • verifiers
  • reward modeling
  • benchmark development
  • evaluation harnesses
  • active learning
  • LangSmith

Pre-training & Fine-Tuning

  • pre-training
  • fine-tuning
  • LoRA/PEFT
  • distributed training
  • data curation
  • PyTorch
  • JAX
  • Hugging Face
  • computer vision
  • Natural Language Processing
  • TensorFlow

Multi-Agentic Workflows

  • multi-agent orchestration
  • agentic workflows
  • Artificial Intelligence agents
  • Model Context Protocol
  • tool calling
  • Large Language Model routing
  • context engineering
  • prompt engineering
  • Retrieval-Augmented Generation
  • embeddings
  • reranking
  • LangGraph
  • LlamaIndex
  • DSPy

Document Intelligence

  • Optical Character Recognition
  • layout detection
  • structured extraction
  • Vision-Language Model extraction
  • schema validation
  • confidence scoring
  • human-in-the-loop review
  • PDF pipelines
  • Tesseract
  • PaddleOCR

Inference / Optimization

  • vLLM
  • TensorRT-LLM
  • SGLang
  • ONNX
  • quantization
  • A100/H100
  • batching
  • KV-cache
  • latency/throughput benchmarking

Machine Learning Operations / Cloud

  • Docker
  • Kubernetes
  • Continuous Integration and Continuous Deployment
  • Metaflow
  • MLflow
  • Weights & Biases
  • observability
  • regression testing
  • GCP
  • AWS
  • FastAPI
  • PostgreSQL

Open source & artifacts.

Evaluation infrastructure, environments, and pipelines built in the open, with merged contributions across the Unstructured ecosystem.

Open source: Unstructured

7 repositories · dozens of merged PRs

Substantial contributions across seven repositories in the Unstructured ecosystem (the leading open-source toolkit for turning unstructured documents into clean, structured data), spanning the full intelligent document processing pipeline, from optical character recognition and layout modeling to application programming interface design and software development kit tooling.

  • Built and optimized parsing pipelines for PDFs, images, emails (EML format), and HTML, with layout-aware extraction for downstream natural language processing
  • Developed and maintained the PaddleOCR and Tesseract wrappers (unstructured.PaddleOCR, unstructured.pytesseract) and backend-agnostic optical character recognition abstractions
  • Contributed to the unstructured application programming interface and unstructured-js-client: RESTful parsing endpoints, consistent element-metadata formats, improved async workflows
  • Maintained Docker base images for reproducible, production-ready optical character recognition and inference deployments

Dozens of merged pull requests across seven repositories, plus issue triage, code reviews, and CI improvements.

Benchmark task authoring

Versioned benchmark tasks with deterministic verifiers and golden references for evaluating coding agents.

Reinforcement learning environment design

Realistic multi-step reinforcement learning environments with reward signals, hidden tests, and failure-mode analysis loops.

Model Context Protocol tools and agent workflows

Typed, auditable Model Context Protocol servers and tool-calling workflows for controlled, inspectable agent execution.

Document artificial intelligence / vision-language model evaluation

Benchmark-driven extraction evaluation with schema validation, confidence scoring, and regression checks.

Production machine learning pipelines

Reproducible training and deployment pipelines with CI, monitoring, and measured inference optimization.

Résumé

Full employment history and credentials, one page.

Download Resume

Nine years of shipped systems.

32 companies, 59 shipped projects behind the case studies above: the full depth, grouped by domain. Company descriptions are public knowledge; engagements without a verifiable public client are shown without company claims.

Artificial Intelligence, Machine Learning & document intelligence

17 engagements

Unstructured

Startup

Builds open-source and commercial document-ingestion tooling that turns PDFs, Word files, and other unstructured documents into structured data for large language model pipelines.

Project

Open-source document intelligence

Parsing, optical character recognition, layout inference, application programming interface delivery, software development kit tooling, and containerized deployment. Merged contributions across 7 repositories, from core inference to the JS client.

  • Python
  • PaddleOCR
  • Tesseract
  • unstructured
  • Docker
  • Representational State Transfer Application Programming Interfaces

Medici Land Governance

Private company

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

Project

Legal and financial document artificial intelligence

Model-routing document pipelines for high-stakes legal and financial documents: optical character recognition, vision-language models, schema enforcement, and cost/accuracy escalation.

  • Vertex AI
  • Gemini
  • Qwen3-VL
  • vLLM
  • Pydantic
  • FastAPI

Sapient Logic

Private company

Builds custom software, cloud, and cybersecurity systems for defense, intelligence, and federal agencies, with a healthcare IT practice under its Araya product line.

Reduced infrastructure cost by 80% in serverless machine learning and application programming interface architecture

4 projects

Security taxonomy mapping

Mapped intrusion-prevention signatures and Common Vulnerabilities and Exposures (CVE) entry descriptions onto MITRE ATT&CK techniques by semantic similarity: two vocabularies describing the same attacks.

  • Word2Vec
  • GloVe
  • TF-IDF
  • Vul_Word2Vec
  • gensim
  • scikit-learn

Araya electronic medical records & natural language processing platform

Electronic medical records server with real Health Level Seven (HL7) message handling (hl7apy) and a React front end, alongside a natural language processing platform sharing JSON Web Token authentication and MongoEngine data models.

  • Python
  • hl7apy (HL7)
  • Flask
  • MongoEngine
  • React
  • Keras

POLAR: offline optical character recognition and translation

Offline document capture and translation for Department of Defense field use: photograph, extract, and translate entirely on the handset. Four runtime-swappable stages and MarianMT pivot routing across 26 language codes, zero network calls.

  • PaddleOCR
  • MarianMT
  • Argos Translate
  • FastText
  • Kivy
  • Kotlin

COMET: Department of Defense intelligence management platform

Led development of COMET, an artificial intelligence-powered intelligence management platform for the Department of Defense: how analysts collect, process, and distribute mission-critical information. Spanning the Electron desktop application for mission workflows and the CFM real-time operations dashboard: a websockets polling service streaming device status to operators with hourly aggregation across 12 subsystems.

  • Electron
  • React
  • Node.js
  • MongoDB
  • Socket.io

DQLabs

Private company

Develops an artificial intelligence-native data quality and observability platform that monitors enterprise data pipelines to detect anomalies and establish data trust.

Trained ten model variants across three axes and kept the full ablation rather than a single headline number

Project

Semantic type detection for personal data discovery

Retrained the Sherlock and Sato research models on a merged corpus to infer 36 semantic column types (Social Security number, employer identification number, bitcoin address, latitude), backstopped by deterministic validators where the answer is decidable.

  • TensorFlow/Keras
  • sentence-transformers
  • PySpark
  • probablepeople
  • geopy

CopyPress

Startup

Operates a fractional content marketing agency producing content strategy, written and visual content, and campaign management for brands and agencies.

Project

Job-title classification by semantic similarity

Classified free-text job titles into nine business categories two ways: a trained neural classifier and nearest-neighbour cosine over SBERT, benchmarked against each other so the trade-off (accuracy vs. absorbing new categories without retraining) was a decision, not a guess.

  • SBERT
  • PyTorch
  • TensorFlow
  • scikit-learn
  • autocorrect
  • pandas

Brandblast AI

Startup

Builds an artificial intelligence platform that automates social-media content creation and scheduling for small businesses and marketing agencies.

Project

Artificial intelligence social media content generation platform

Platform that generates and publishes tailored social posts by learning a business's unique brand voice: GPT-3 generation fine-tuned for social content, and a brand-analysis engine that extracts identity, tone, and style markers from a company's website.

  • GPT-3
  • Large Language Model
  • Brand analysis engine

Averi AI

Startup

Builds an artificial intelligence marketing workspace that carries content from strategy to publishing, combining artificial intelligence models with a vetted expert network.

Project

Generative artificial intelligence marketing copilot platform

Full-stack build of Averi's marketing copilot: OpenAI and Together.ai models behind content generation and optimization pipelines, on a server-rendered Next.js architecture.

  • OpenAI
  • Together.ai
  • Next.js
  • Large Language Model

SpeechLab

Startup

Builds speech-to-speech artificial intelligence translation and dubbing tools that localize video and audio into 50+ languages while preserving the original speaker's voice.

475 integration-level test assertions on the application programming interface layer

Project

Speech translation and dubbing platform

Automatic speech recognition → machine translation → text-to-speech as a product: Whisper recognition, pyannote diarization, NeMo speech modelling, DeepL translation, and Ray distributed execution around a five-stage corpus pipeline, a JSON Web Token-authenticated Node application programming interface, a Next.js front end, and Terraform infrastructure.

  • Whisper
  • pyannote
  • NeMo
  • Ray
  • Node/Express
  • Next.js
  • Terraform

Foundations.

Formal training behind the systems above.

Education

  • B.A. in Computer Science

    UC Berkeley

  • B.A. in Cognitive Science

    UC Berkeley

Certifications

  • Machine Learning SpecializationStanford University
  • Deep Learning SpecializationDeepLearning.AI
  • IBM Data Science SpecializationIBM
  • AWS Cloud Practitioner EssentialsAWS
  • Certified Scrum Product Owner (CSPO)Scrum Alliance
  • Google Business Intelligence SpecializationGoogle
  • Google Data Analytics SpecializationGoogle

Frequently Asked Questions

What roles is Christine open to?

Open to remote Reinforcement Learning, Post-Training, and Agentic Artificial Intelligence roles.

What is her core expertise?

Reinforcement learning (supervised fine-tuning, direct preference optimization, and group relative policy optimization, plus environment design and deterministic verifiers), computer vision (production detection and optical character recognition/vision-language model extraction at 10M+ images/day), and multi-agentic systems.

Where has she shipped production artificial intelligence?

Across defense (classified Department of Defense programs), healthcare (healthcare privacy law-regulated environments), and enterprise software, with merged contributions across 7 repositories in the Unstructured open-source ecosystem.

What is her background?

UC Berkeley, B.A. in Computer Science and B.A. in Cognitive Science; 9+ years shipping production artificial intelligence and machine learning systems.

Contact

Training or fine-tuning models, building reinforcement learning environments, or designing multi-agentic workflows? Let’s talk.

Open to remote Reinforcement Learning, Post-Training, and Agentic Artificial Intelligence roles

Location
Remote / United States
Email MeLinkedInGitHubNotion PortfolioOnline ResumeDownload Resume