AI Purchase Invoice Entry for Paladiya Construction

The situation

We had already built the client’s expense management app: clients, sites, labour, materials, suppliers, payments received and site-wise reports.

Purchase invoices were the slow part. Every supplier bill for cement, steel or sand had to be typed in line by line - material, HSN code, quantity, rate - with CGST, SGST or IGST worked out for each line.

What we built

An AI purchase invoice flow in three steps: upload, review, confirm.

• Upload: staff select one or many invoices as PDF, JPG, PNG or WEBP. Multi-page PDFs are read page by page.
• Extract: a vision model reads the document and returns supplier name, phone, address and GSTIN, the invoice number and date, the tax type and every line item.
• Each line carries the material, HSN code, quantity, unit price, GST rate, the CGST, SGST and IGST amounts and the line total.
• Review: the extracted data opens beside a preview of the original document. Staff correct anything before it is saved.
• Confirm: the invoice is recorded against a site. The supplier and materials are matched to existing records or created, without duplicates, and the material expenses are posted.

Checks around the AI

The model’s output is held to a fixed schema, then validated and cleaned before anyone sees it: phone numbers, GSTINs, dates and field lengths. A file that is not an invoice is rejected with a clear message instead of creating bad records.

Nothing reaches the accounts until a person has reviewed and confirmed it.

The result

Because every line now carries its GST split, the same data drives the app’s monthly GST report and a list of invoices unpaid for more than 45 days.

The same approach was extended to subcontractor running account bills, where the AI also reads retention and other deductions.