Build an Automatic AI Accountant With Muse, Unsiloed, and Jev

Description and use

An automated bookkeeper running in Muse uses Unsiloed to read documents and Jev to make classification decisions, enabling small businesses to maintain books without a human accountant.

Problem

Small businesses need a way to reconcile receipts, bills, payouts, and bank statements into a ledger without a human accountant.

Outcome

The system processed eight documents for a glassblowing studio, produced a reconciled ledger, and delivered a month‑end pack with a profit of $2,194.47, correctly categorizing a $6,670 oven as equipment and splitting a payout into gross income and platform fee.

Workflow

• Owner attaches month’s documents to Muse
• Muse calls Unsiloed /v2/extract for each document type
• Unsiloed returns fields with confidence scores and citations
• Python code validates totals, bank balances, asset thresholds, duplicates, and bank matches
• Jev model answers narrow yes/no or choice questions about account classification, personal vs business, asset status, duplicates, and payments
• Muse aggregates results, books entries, and saves owner responses as rules for future months
• Muse generates month‑end pack (profit & loss, assets, liabilities, bank reconciliation) and sends to owner

Evidence

In the age of AI, a small business shouldn't need an accountant to keep its books.

Source

It asks the owner only about the items it can't settle, in plain English.

Source

Unsiloed’s /v2/extract endpoint takes a JSON schema for each document type and returns every field with a confidence score and, with enable_citations on, a bounding box on the page.

Source

Jev judges only what's left. Code decides that the annealing oven costs more than $2,500, and Jev decides whether it's durable equipment.

Source