I ask "what if" when a process feels manual, then build the prototype. This is the product, design, and automation work behind that instinct — from AI-powered data migration architecture to the tools built to prove it out.
A five-stage pipeline that parses payroll/HRIS files from any format or provider, classifies every earning and deduction code by tax treatment, reasons through multi-state withholding obligations, and produces platform-ready setup instructions — replacing manual, template-per-client mapping with a system that reasons through the tax logic itself.
PDF table extraction, Excel multi-sheet ingestion, CSV parsing. A deterministic pre-screener identifies the provider by filename, headers, and body text. High-confidence matches load the right parsing instructions directly — no LLM call needed.
Maps every earning and deduction code to a tax-treatment taxonomy — six earning types, three deduction types — with per-code tax flags and confidence scores. Low-confidence codes flag for human review.
Applies a payroll tax knowledge base: multi-state withholding rules, wage splits by days worked, SUI cap tracking per state, and special-case overrides. Produces a full recommendation with human-readable reasoning behind each decision.
Deterministic business rules confirm gross minus deductions minus taxes equals net (within $0.02), that no employer taxes appear in employee output, that YTD values aren't mistaken for current-period figures, and that reimbursements aren't miscategorized as gross pay. Failures route to a review queue.
Produces platform-ready output with per-item setup instructions — earning code configuration, deduction tax flags, state withholding setup, SUI wage recording — ready for import.
A deterministic pre-screener scores provider fingerprints and returns a confidence score. High confidence loads the right parsing instructions directly, skipping the LLM step entirely. For known, high-volume providers, AI cost per file approaches zero — and adding a new provider means new instructions, not new code.
A full 55-jurisdiction reciprocity matrix determines which state withholds, whether both states require wage reporting, whether wage splitting applies, and whether a special case exists — for example, KY/VA reciprocity applies only to daily commuters, and Maryland residents working in DC owe both DC income tax and Maryland's local tax.
Handles the distinctions responsible for the most common silent errors in payroll implementation — for example, a traditional 401(k) reduces federal and state income tax but not FICA, FUTA, or SUTA, while imputed income from group term life is income/FICA-taxable but FUTA-exempt. Unknown codes default conservatively and flag for review.
The underlying tax rules — SUI wage bases, reciprocity agreements, federal constants — are stored as versioned data, not hardcoded logic. Annual updates are data changes, not code changes.
A full self-service payroll import platform, built end-to-end as a Next.js prototype to test whether a guided, wizard-style import experience could reduce dependency on specialist-led setup.
A working, persistent (localStorage-backed) prototype showing that complex payroll import — normally requiring a trained specialist — could be guided step by step for a broader range of users, informing later product conversations about self-service tooling.
A recurring "what if this didn't require manual entry" question, turned into a shipped workflow redesign.
Championed and helped launch implementation and processing for international payroll, partnering directly with development, product, and banking teams to define SOPs for a market with materially different compliance requirements than U.S. payroll.
Two connected efforts: getting new hires to full performance faster, and helping an entire implementation org build real fluency with AI tools.
Redesigned and deployed a successful New-Hire Training Program that resulted in new hires hitting 12-month performance benchmarks by month 6.
Created and facilitated multiple soft skill development courses, including successful communication, ownership and accountability, project management, persuasion and influence, and prioritization and time management.
Responsible for the AI and Automation Roadmap in implementation, including the training program. Created and facilitated "AI for Implementation" coursework required for the entire department, focused on practical AI use and reframing AI as a tool for better decisions, not fewer jobs.