Senior Machine Learning Engineer

Medici Land Governance Inc.; United States; On-site; USD 150,000 - 210,000; Full time

Job summary

Medici Land Governance Inc. is hiring a hands-on technical leader to own our end-to-end document intelligence platform. You will own the architecture of document intake, quality assessment, OCR/VLM extraction, indexing, and serving strategies for batch and real-time workloads.

The role emphasizes reducing hallucinations, improving production accuracy, and building resilient systems in a fast-changing model landscape while leading a small AI engineering team.

Qualifications

• 6+ years building production AI/ML systems.
• Strong experience in document AI, OCR, NLP, or information extraction.
• Experience owning or improving intelligent document processing pipelines.
• Experience evaluating multiple model providers and APIs in production.
• Hands-on with vLLM, Triton, or similar inference-serving systems.

Responsibilities

• Architect and improve document AI pipelines for court and property records.
• Build systems to assess image quality on arrival and route processing paths.
• Optimize preprocessing for scan quality and alignment.
• Design robust fallback across multiple model providers to reduce vendor lock-in.
• Evaluate platforms for resiliency, cost, and flexibility.
• Develop eval frameworks with regression tests and versioning.

Skills

Document AI

OCR

LLM infra

Python

PyTorch

MLOps

System design

Education

BS in CS or equivalent

Tools

vLLM

Triton

Textract

MLflow

Langfuse

Job description

We're hiring a hands-on technical leader to own our end-to-end document intelligence platform.

Our system processes court documents, property records, and other complex scanned files across both clean new recordings and poor-quality historical documents. We need someone who can lead architecture and execution across document quality assessment, preprocessing, OCR/VLM extraction, indexing, provider fallback, evals, and serving strategy.

This is a high-ownership role for someone who can make strong technical decisions, improve production accuracy, reduce hallucinations, and build resilient systems in a fast-changing model landscape.

What you'll own

• document intake, quality assessment, and routing
• image preprocessing and scan enhancement
• OCR, VLM, and document extraction workflows
• structured extraction and indexing
• fallback across multiple AI providers and standalone OCR/model systems
• evals, regression testing, and prompt/model versioning
• serving strategy for batch and real-time workloads
• latency, throughput, and memory optimization
• confidence scoring, validation, and hallucination reduction

What you'll do

• Architect and improve document AI pipelines for court records, property records, and other difficult document sets
• Build systems that assess image quality and document complexity on arrival and determine the correct processing path
• Optimize preprocessing for skew correction, denoising, deblurring, contrast enhancement, binarization, orientation correction, cropping, and scan recovery
• Design robust fallback across multiple paid model providers and standalone OCR/VLM systems to reduce vendor dependency
• Evaluate platforms such as direct APIs, OpenRouter, Vertex AI, and self-hosted inference for resiliency, cost, and operational flexibility
• Assess paid API vs self-hosted economics, including infrastructure cost, memory usage, throughput, fine-tuning needs, and long-term maintainability
• Build or optimize self-hosted serving systems using vLLM or comparable inference stacks
• Support both offline batch workloads and lower-latency live endpoints, including queueing and prioritization tradeoffs
• Improve compatibility between OCR outputs, custom internal formats, and downstream indexing/extraction systems
• Build an evals framework using MLflow or equivalent tooling, including ground-truth datasets, regression testing, prompt versioning, and user-feedback integration
• Establish measurable standards for OCR accuracy, extraction accuracy, hallucination rate, indexing quality, latency, throughput, and cost per document
• Lead failure analysis and iterative improvement, especially for degraded historical records
• Provide technical leadership to a small AI engineering team while remaining deeply hands-on

What we're looking for

• 6+ years building production AI/ML systems
• Strong experience in document AI, OCR, computer vision, NLP, or information extraction
• Experience owning or materially improving intelligent document processing pipelines
• Experience evaluating multiple model providers, APIs, and document AI services in production
• Strong understanding of paid APIs vs open-source/self-hosted tradeoffs
• Hands-on experience with vLLM, Triton, TGI, or similar inference-serving systems
• Experience with tools such as Tesseract, PaddleOCR, Textract, Azure Document Intelligence, Google Document AI, LayoutLM, Donut, or similar
• Strong Python skills and experience with PyTorch, OpenCV, Hugging Face, and modern ML infra
• Experience building eval frameworks with regression testing, ground truth, versioning, and production monitoring
• Strong understanding of batching, scheduling, GPU memory, latency, throughput, and cost optimization
• Experience reducing hallucinations through validation, confidence scoring, schema enforcement, and guardrails
• Ability to lead technical direction and drive execution in an ambiguous environment
• Court records, property records, legal documents, or public-records experience
• Historical document digitization or degraded scan handling
• Hybrid OCR + VLM + rules + human-review system design
• ML/LLM observability using MLflow, Langfuse, W&B, or similar
• Priority scheduling, queue design, and mixed batch/live inference systems
• Fine-tuning document models when vendor APIs are insufficient
• Experience leading a small team or owning a critical AI platform

Who thrives here

• Someone who wants real ownership
• Someone who can make decisions and move
• Someone who is hands on
• Someone who likes messy data and operational complexity
• Someone who cares about accuracy, resiliency, speed, and cost
• Someone who can lead without hiding behind process

Who is not a fit

• Someone looking for a pure research role
• Someone who only wants prompt engineering or prototyping
• Someone who has not owned production systems with reliability and cost constraints
• Someone who prefers narrow model work over full-platform responsibility