I build AI-powered finance systems that turn messy financial data into accurate, audit-ready answers — and keep accounting, payments, and CRM systems in perfect sync.
- 🔭 I build end-to-end accounting & payments integrations (QuickBooks Online, Xero, Stripe) synced into a single source of truth
- 🤖 I build AI finance assistants — RAG over invoices, contracts, statements & GL data with cited, traceable answers
- 📄 I handle financial document ingestion — invoices, receipts, bank statements, W-9/1099s → structured data with OCR fallback
- 🔁 I ship idempotent syncs, reliable webhooks, and reconciliation engines that never double-post
- 🎯 My north star: every figure traceable to its source document, line item, or transaction ID
Accounting / Payments
AI / LLM / RAG
Vector DBs / Embeddings
Backend
Frontend
Infra / DevOps
Full-stack web app that automates invoicing for a title-and-registration business, keeping three systems — QuickBooks Online, a GraphQL CRM, and the client's internal app — in perfect sync.
- Ingests messy vendor batch files (PDF / Excel / CSV) → unified invoice schema with LLM-assisted extraction
- Template-matching engine auto-generates QuickBooks invoices
- Two-way payment sync via webhooks — payments, refunds & status updates flow bidirectionally
- Customer de-duplication + idempotent sync layer → zero duplicates across systems
Next.js (App Router) · React 19 · TypeScript · Azure SQL · QuickBooks Online API · GraphQL/REST · OAuth 2.0
Cited, audit-ready Q&A over 50k+ invoices & statements with full source traceability and
<2sp95 latency.
Python · FastAPI · LlamaIndex · GPT-4o · Qdrant · PostgreSQL
Auto-matched 120k+ monthly transactions at 99.2% accuracy, cutting manual review by 90%.
Python · Stripe Webhooks · QuickBooks Online API · idempotent sync
| Project | Scale | Outcome |
|---|---|---|
| QuickBooks + Stripe reconciliation | 120k+ tx / month | 99.2% auto-match, −90% manual review |
| AI finance Q&A | 50k invoices & statements | Cited answers, <2s p95, full audit trail |
| Xero AP automation | 18k invoices / month | 98.7% parse success via OCR fallback |
Send me your stack (QuickBooks / Xero / Stripe / Servis.ai), data volume, and the accuracy bar you need to hit — I'll reply with an integration approach and the two or three risks I'd tackle first.
⭐️ Every number traceable. Every sync idempotent. Every answer cited.