An AI-assisted pipeline that reads invoices, matches vendors, and posts GST-compliant draft bills to Zoho Books under human approval.
Accounts payable work in India still runs largely on manual entry: read the invoice, retype it into the accounting system, apply the GST rules by hand. Formats vary, vendor names never quite match, and there is rarely a record of who touched what. Our client wanted that effort removed without giving up control over what reaches their books.
We designed the workflow around one decision: AI reads, a person approves. That boundary shaped the architecture. An OpenAI vision model extracts vendor details, line items, taxes, and dates from uploaded PDFs and images, and a reviewer confirms the result in a side-by-side interface before anything is published.
Vendor identification needed more than a name lookup, so we built a cascade that tries GSTIN first, then phone, then exact name, then fuzzy matching with RapidFuzz, and finally LLM disambiguation. An unmatched vendor is created at publish time. A second LLM pass picks the GL account from vendor and line-item context.
Every bill lands in Zoho Books as a draft with the original document attached, which keeps final posting in the client’s hands. Each published invoice is written to Google Sheets by year and month, so the finance team has an audit trail they can open without us.
Full automation would have put unverified data into the client's books. Approval was placed at the end of every run, which made the AI useful without making it the final authority.
Each invoice follows the same fixed path from upload to audit log. Finance teams need to know what the system will do with any document, and predictable handling is what earns that trust.
Name matching alone produced duplicates. A cascade from GSTIN through fuzzy matching to LLM disambiguation resolved vendors accurately and kept the ledger clean.
Bills reach Zoho Books as drafts, and final posting stays a human action. That single constraint keeps an extraction error from landing straight in the accounts.
Audit records were written to Google Sheets, organized by year and month. Finance staff can open, filter and share that log without a developer or a custom screen.
What Makes Us The Buzz of Tech Town?
Contact Us
Have Questions? Let's Talk!
Fill out this form to drop us an email, and we will reach out to shape something extraordinary.