AI Engineer

Stealth Startup YC backed; Chicago (IL); On-site; USD 150,000 - 200,000; Full time

Job summary

Stealth Startup YC backed in Chicago seeks an AI engineer to own the intelligence layer across the document-to-return workflow. You will design production AI systems that convert messy documents into machine-readable tax data and support tax professionals with confidence signals and source citations.

Join a small, high-ownership team building the core tax engine and interfaces, collaborating with tax professionals and engineers to improve accuracy, speed, and auditability in real-world filings.

Qualifications

• Experience building robust OCR and extraction for PDFs, forms, and scanned docs.
• Design scalable evaluation frameworks with ground-truth labels and metrics.
• Strong judgment on observability, reproducibility, and AI-in-the-loop workflows.
• Bias toward shipping real-world document AI with auditability and reliability.

Responsibilities

• Build systems that classify client documents and extract reliable tax data.
• Design end-to-end workflows to extract fields and resolve ambiguity.
• Develop agents to compare a completed return with source documents and suggest corrections.
• Define confidence signals, citations, and review workflows.
• Establish observability and feedback to measure QA and outcomes.

Skills

Document OCR

LLM‑based Document Understanding

Structured Data Extraction

Evaluation Framework Design

Ground-Truth Labeling

Statistical Analysis

Agentic Systems

Human-in-the-Loop Workflows

Observability

Reproducibility

Production AI Systems

Python

Job description

Our client is building the modern tax engine and application that makes professional tax preparation faster, more reliable, and easier for tax professionals. The company was founded by Harvard Business School Alums backed by YCombinator and PeakXV.

About the Role

Our client is seeking an AI engineer to own the intelligence layer across the document-to-return workflow. You will design and maintain systems that turn messy client documents into usable tax data, help tax professionals identify missing or inconsistent information, and review completed returns before filing. This is not a generic chat-experience role — you will build production AI systems that operate on real financial documents and work alongside a deterministic tax engine, where accuracy, traceability, and thoughtful human control matter.

Responsibilities

• Build systems that collect client documents and classify them, including tax forms, financial statements, receipts, and supporting materials.
• Turn structured and unstructured documents into reliable, usable data — designing workflows to extract fields, preserve source context, resolve ambiguity, and route low‑confidence results for human‑in‑the‑loop review.
• Build agents that examine a completed return alongside its source documents and relevant tax context, identifying missing information, inconsistencies, and potential errors, then explain findings clearly and propose corrections for a tax professional to review.
• Shape how tax professionals understand, trust, and act on AI output, including confidence signals, citations to source documents, review workflows, correction flows, and clear boundaries between suggestions and final decisions.
• Build the evaluation, observability, and feedback systems needed to measure extraction quality, agent performance, and user outcomes over time.
• Understand and optimize the document and return‑preparation workflow, ship meaningful improvements, expand extraction coverage, and pinpoint the most time‑consuming manual steps where AI can improve accuracy, speed, or consistency.
• Own a production workflow for document classification and extraction with clear quality metrics and a feedback loop from users.
• Establish the foundation for expanding AI across the application into an AI‑powered review experience that helps tax professionals catch issues earlier and prepare returns with greater confidence.

Qualifications

• Experience building robust document OCR and extraction systems, including working with PDFs, forms, scanned documents, and financial data, and applying modern LLM‑based methods for document understanding, parsing, and structured data extraction.
• Experience designing scalable evaluation frameworks that reliably benchmark quality, including defining evaluation datasets, ground‑truth labels, scoring methodologies, and statistical analyses for structured and unstructured documents.
• Strong judgment about evaluation design, observability, reproducibility, and human‑in‑the‑loop workflows.
• A bias toward shipping and learning from real‑world document and model behavior, paired with a high bar for accuracy, auditability, reliability, and actionable measurement.

Required Skills

• Document OCR & Extraction
• LLM‑based Document Understanding
• Structured Data Extraction
• Evaluation Framework Designli>
• Ground-Truth Labeling
• Statistical Analysis
• Agentic Systems
• Human-in-the-Loop Workflows
• Observability
• Reproducibility
• Production AI Systems
• Python

Pay range and compensation package

• $150 - $200K
• High ownership with little separation between product and engineering.
• Direct collaboration with tax professionals and the engineers building the core tax engine and application.
• Consequential decisions about how AI becomes part of a new tax platform, on a small, high‑ownership team.