ReconAI: AI-Powered Bookkeeping Reconciliation

Bookkeepers spend 8-12 hours per week manually matching bank transactions to invoices across statements, payment apps like GCash and Maya, and supplier records. It's tedious and error-prone — an estimated 67% of small businesses have accidentally made duplicate payments, and weak audit trails make financial disputes hard to prove. ReconAI automates this workflow. It takes bank transactions and supplier invoices, then uses an LLM via Fireworks AI (running on AMD hardware) to intelligently match them — even when amounts differ slightly due to transaction fees, or dates are off by a day. Simple rule-based matching fails in these cases; an LLM can reason through the ambiguity and explain its decision in plain language. Beyond matching, ReconAI flags duplicate transactions and transactions missing supporting invoices — the exact problems that eat hours of a bookkeeper's week. Every reconciled record is sealed into a SHA-256 hash chain, where each entry's hash includes the previous entry's hash. This makes the ledger tamper-evident: any edit after the fact breaks the chain and is immediately detectable, without needing blockchain infrastructure. The MVP uses realistic Philippine business demo data (GCash, Maya, Meralco, PLDT, and more) to demonstrate the full pipeline: 20 transactions, 17 invoices, automatic matching, duplicate/mismatch/missing-documentation flagging, and a live hash chain visualizer — all running through a FastAPI backend, fully containerized with Docker. Built for bookkeepers and small business owners drowning in fragmented, multi-channel financial records — not as another generic AI chatbot, but as a focused tool solving one real, painful workflow.