Lead AI Engineer - Audit Automation

We are a specialized CPA firm that runs high-stakes employee benefit plan audits and we are hiring the first engineer to rebuild our entire audit workflow from the inside out. If you want to own a greenfield AI platform end-to-end, work directly with the practitioners who use it every day, and see your code replace thousands of hours of manual rekeying, this is that role. What you'll do
Map and automate the audit lifecycle: embed with audit partners and senior managers to understand exactly how work flows, then design and ship intelligent agents that eliminate manual steps one by one.
Build a governed data layer: every value written by an AI agent carries its source, author, verification status, and timestamp so a partner can inspect and sign off with confidence.
Develop automated roll-forward: generate current-year workpapers from prior-year files automatically, flagging year-over-year changes for human review instead of requiring manual retyping.
Create intelligent document intake: extract structured records from client-supplied and third-party files; auto-draft follow-up questions when data is missing or contradicts historical records.
Render results into standard templates: write automated output back into the firm's Excel and Word workpapers with full visual and structural fidelity; lock records securely on final archive.
Own production operations: access control, model policy enforcement, cost monitoring, incident response, and parallel-run discipline before any automated step displaces a manual one.
Set reusable patterns: a solution built for one client becomes the baseline configuration for all others, compounding your impact across the practice.
What you need
Production Python service ownership: you have been the accountable engineer for a live, typed, async backend service and can walk through real failures and how you resolved them. Strong TypeScript engineers with a compelling stack-alignment case will also be considered.
LLM-backed workflow engineering: you have shipped systems where LLM outputs are captured as strictly structured data (Pydantic, Instructor, or native structured-output APIs), validated before ingestion, and wrapped in deterministic logic.
Agent and tool-calling experience: practical production deployment of agents, tool-calling, or MCP integrations with measurable accuracy tracking.
Data boundary judgment: you have made deliberate, defensible decisions about how sensitive data is isolated before it reaches external APIs; you understand zero-data-retention agreements and regional endpoint configurations.
Relational data modeling: strong PostgreSQL experience and comfort with migration frameworks (SQLAlchemy/Alembic); you think of the platform as an authoritative data model first, application layer second.
Nice to have
Programmatic Office automation: generating formatted Excel and Word files from code (openpyxl, python-docx, Office.js) including formulas, named ranges, and sheet protection.
Cloud infrastructure on Azure: AKS, App Service, Blob Storage, Azure OpenAI endpoint management.
Enterprise agent tooling: Azure AI Foundry, Copilot Studio, Power Platform, or Dataverse.
Async pipeline orchestration via task queues (taskiq, RabbitMQ, Redis) or state machines.
Continuous LLM evaluation: custom regression suites that track output quality as models or prompts change.
Cross-functional rollout leadership with non-technical subject matter experts resulting in sustained adoption.
Domain knowledge in audit, accounting, compliance, or payroll.
Document extraction experience (Azure Document Intelligence, OCR/document-understanding platforms).
Modern observability (OpenTelemetry to cloud monitoring suites).
<div class="otQkpb" style="border-bottom-color:rgb(1

Work arrangement
Yes