Applied Machine Learning Engineer

Lead Applied Machine Learning Engineer – Document Intelligence Company: Medici Land Governance Employment Type: Full-Time Level: Lead / Principal About the Role Medici Land Governance is looking for a Lead Applied Machine Learning Engineer to own and advance our document intelligence capabilities for property records, legal documents, and public-record data. Our platform processes large and diverse document collections ranging from clean modern recordings to decades-old historical records with degraded scans, complex layouts, handwriting, stamps, legal descriptions, tables, and inconsistent formatting. We are expanding our use of proprietary machine learning models and domain-specific training data to improve how these documents are classified, restored, read, understood, and converted into structured information. This role will lead the technical direction for our document ML systems, including model development, fine-tuning, training data strategy, evaluation, computer vision, OCR, multimodal models, and production inference. This is a deeply hands-on role for someone who wants to own a difficult real-world machine learning problem from data and experimentation through production deployment and continuous improvement. What You’ll Own Technical direction for our document intelligence and applied ML systems Proprietary and domain-specialized document models Training, validation, and ground-truth datasets Document classification and quality assessment Image preprocessing and document restoration OCR and text recognition Layout and document structure understanding Structured information extraction Vision-language and multimodal model evaluation Fine-tuning and model adaptation Confidence scoring and validation Model evaluation and regression testing Production ML inference and optimization Continuous improvement using production failures and feedback What You’ll Do Develop and improve machine learning models specialized for property records and public-record documents. Fine-tune and adapt open-source computer vision, OCR, transformer, and vision-language models using domain-specific datasets. Evaluate model architectures and determine when to fine-tune, distill, combine, or develop specialized models. Improve document classification, OCR, layout understanding, and structured extraction accuracy. Build systems capable of handling both clean modern recordings and difficult historical records. Develop preprocessing strategies for skew correction, denoising, deblurring, contrast enhancement, orientation correction, cropping, binarization, and degraded scan recovery. Improve recognition of faded text, handwriting, stamps, seals, unusual fonts, damaged pages, poor microfilm, and inconsistent layouts. Develop structured extraction for document types, parties, recording information, parcel data, legal descriptions, mortgages, liens, releases, and other property-record information. Design and maintain high-quality training, validation, and ground-truth datasets. Build feedback loops that convert production failures and human corrections into improved training data. Develop confidence scoring and validation systems that reduce incorrect or hallucinated results. Build rigorous evaluation and regression-testing frameworks for model and pipeline changes. Establish measurable standards for OCR accuracy, field extraction, document classification, hallucination rate, document-level accuracy, latency, throughput, and cost. Deploy and optimize models for high-volume production inference. Optimize GPU utilization, batching, memory consumption, concurrency, latency, and throughput. Benchmark proprietary models against commercial OCR, document AI platforms, and leading multimodal models. Provide technical leadership to a small AI/ML engineering team while remaining highly hands-on. What We’re Looking For 6+ years building production machine learning or applied AI systems. Strong experience in computer vision, OCR, document AI, multimodal ML, NLP, or information extraction . Experience training or fine-tuning machine learning models rather than exclusively integrating commercial AI APIs. Strong Python and PyTorch experience. Experience developing and curating training and ground-truth datasets. Strong understanding of transformer-based vision, language, or multimodal architectures. Experience with model evaluation, error analysis, experimentation, and regression testing. Experience deploying ML models into production. Understanding of GPU inference, memory utilization, batching, latency, throughput, and model optimization. Ability to analyze model failures and systematically improve performance. Ability to establish technical direction and drive execution in an ambiguous environment. Relevant Technologies Experience with some of the following is valuable: PyTorch, Hugging Face, OpenCV, LayoutLM, Donut, TrOCR, PaddleOCR, DocTR, Tesseract, Vision Transformers, Vision-Language Models, MLflow, Weights & Biases, vLLM, Triton, ONNX, TensorRT. Experience with every technology listed is not required. Strong Bonus Points Property records, deeds, mortgages, liens, title records, court records, legal documents, or public-record experience. Historical document digitization. Degraded scan restoration. Handwriting recognition. Experience developing proprietary OCR or document-understanding models. Fine-tuning multimodal or vision-language models on proprietary datasets. Synthetic training-data generation. Active learning or human-in-the-loop ML systems. Model quantization, distillation, or optimization. Self-hosted GPU inference. Open-source contributions or research in computer vision, OCR, multimodal ML, or document intelligence. Who Thrives Here You’ll likely thrive in this role if you: Want to develop and improve models, not simply call APIs. Enjoy solving difficult problems with messy real-world data. Care deeply about measurable accuracy. Can combine classical computer vision with modern deep learning. Are comfortable moving between experimentation and production. Like investigating why individual documents fail. Can make strong technical decisions and move quickly. Want ownership of a critical machine learning platform. Can lead technically while remaining hands-on. Who Is Not a Fit This role is unlikely to be the right fit if your experience is primarily: Prompt engineering ChatGPT wrappers RAG applications without underlying model development LangChain or agent development without deeper ML experience Commercial AI API integration without model training or evaluation Pure MLOps or infrastructure Pure data engineering Academic research without production deployment The Opportunity Property records present an unusually challenging document intelligence problem because quality, structure, terminology, and formatting can vary dramatically across jurisdictions and decades. This role will have significant ownership over how Medici applies machine learning to that problem and will help develop increasingly specialized models, datasets, and evaluation systems around our domain. The objective is a document intelligence platform that becomes more accurate, resilient, efficient, and specialized as it processes more of the records we work with.