Machine Learning
ML Engineer, Document AI
Own the models that read county assessment notices, rent rolls and operating statements and turn them into structured data a consultant can trust. Every field you extract ends up in an evidence packet that someone presents to a county board.
What you will do
- Build and improve OCR and layout-aware extraction for assessment notices, whose formats differ in every county and change from year to year.
- Extract tables from rent rolls and operating statements that arrive in many layouts, including exports from Yardi Voyager, MRI, RealPage and Excel.
- Build evaluation harnesses that report field-level precision and recall by jurisdiction and document type, and block releases that regress.
- Design confidence scoring and human review queues so low-confidence fields reach a person before they reach a petition.
- Train and refine models only on public records, licensed data and synthetic documents. Customer property financials are never used for training.
- Monitor extraction quality in production and trace each error to its cause, whether the scan, the layout or the model.
What you bring
- 4 or more years of applied machine learning, with at least 2 years on document understanding, OCR or information extraction.
- Strong Python and PyTorch, and experience taking a model from notebook to monitored production service.
- Rigor with evaluation: labeled test sets, error analysis and metrics that reflect the cost of a wrong field.
- Ability to explain a model's output in plain terms to non-engineers, including appraisers and consultants.
Nice to have
- Experience with financial or legal documents, such as leases, statements or tax forms.
- Experience generating synthetic documents for training and testing.
- Familiarity with property records, assessment notices or county data.
- Experience with active learning or review tooling.