Azure AI Document Intelligence Engineer

Role: Senior Azure AI Document Intelligence Engineer 

Location: Edina, MN/ Chicago, IL/ Irving, TX

Long Term

Type: Contract

Job Description:

Senior Azure AI Document Intelligence Engineer design, train, evaluate, and continuously improve the document-processing models used for supply-chain contracts and invoices.
This engineer will own extraction quality and be directly accountable for achieving the project’s 98–99% accuracy objectives.

Key Responsibilities

• Analyze invoice and contract variations across suppliers, countries, languages, formats, and scan-quality levels.
• Define the document taxonomy and extraction schema for:
• Invoice headers
• Contract metadata
• Purchase-order references
• Supplier and customer information
• Dates, currencies, taxes, discounts, and totals
• Payment terms, renewal dates, obligations, and termination clauses
• Tables and invoice line items
• Evaluate Azure AI Document Intelligence prebuilt invoice, layout, custom neural, custom template, classification, and composed-model capabilities.
• Build and train custom document classifiers and extraction models.
• Prepare, label, clean, balance, and version training and evaluation datasets.
• Implement document preprocessing for rotation, skew, noise, resolution, page separation, and scan-quality issues.
• Develop confidence-scoring and validation strategies using field-level and OCR-level confidence signals.
• Implement deterministic validation rules, including:
• Subtotal, tax, and total reconciliation
• Currency and date validation
• Purchase-order and supplier matching
• Duplicate-document detection
• Required-field and cross-field consistency checks
• Establish confidence thresholds and human-review rules for uncertain extractions.
• Perform error analysis by document type, supplier, field, language, and scan quality.
• Create automated evaluation pipelines reporting exact-match accuracy, precision, recall, F1 score, false-positive rates, and false-negative rates.
• Monitor model drift and retrain models when document formats or business requirements change.
• Document model versions, training data, experiments, limitations, and release decisions.
• Collaborate with supply-chain, procurement, accounts-payable, legal, and engineering teams.

Required Qualifications

• Bachelor’s or master’s degree in Computer Science, Data Science, Artificial Intelligence, Engineering, or a related discipline.
• 4+ years of experience in machine learning, OCR, computer vision, NLP, or intelligent document processing.
• Hands-on experience with Azure AI Document Intelligence or a comparable enterprise document-processing platform.
• Strong Python development skills.
• Experience with REST APIs, JSON, Azure SDKs, and asynchronous processing.
• Experience training and evaluating document classification and field-extraction models.
• Strong understanding of accuracy, precision, recall, F1 score, confidence calibration, and test-set design.
• Experience extracting complex tables and variable-length line items.
• Experience creating labeled datasets and maintaining data-quality standards.
• Familiarity with invoices, purchase orders, contracts, or supply-chain documents.
• Strong analytical, debugging, and technical-documentation skills.

Preferred Qualifications

• Experience with Azure Machine Learning, MLflow, Azure AI Search, or Azure OpenAI.
• Experience with multilingual documents.
• Familiarity with contract clause extraction and legal-document processing.
• Knowledge of ERP or procurement systems such as SAP, Oracle, Dynamics 365, Coupa, or Ariba.
• Experience implementing active learning, model-drift monitoring, and human-in-the-loop workflows.
• Knowledge of data privacy and document-retention requirements.
• Success Measures
• ≥99% exact-match accuracy for approved critical fields.
• ≥98% exact-match accuracy across all agreed fields.
• Accuracy maintained across suppliers, formats, and document-quality categories.
• Measurable reduction in manual corrections and review time.
• Documented improvement process for fields that miss their thresholds.
• No production model deployed without passing the approved evaluation suite.