Skip to main content

DataFlorent

InvoiceX

Multi-tenant web platform that reads supplier invoices, verifies them and gets them ready for the accounting books — with a human reviewer always in control.

THE PROBLEM AND THE OBJECTIVE

Finance teams in Indian SMEs and CA firms still re-key purchase invoices by hand: reading PDFs and scans, checking the supplier’s GSTIN, re-adding totals and typing vouchers into Tally. It is slow, error-prone and leaves no audit trail. Many AI tools extract data but behave like a black box, so accountants cannot tell when to trust them. Objective: build a self-serve product that turns an uploaded invoice into verified, export-ready data in under a minute, explains every doubt it has, and keeps a person in charge of the final decision — sold on a free trial with a simple subscription.

MAIN FEATURES

  • Vault upload & AI extraction — batch upload of PDF/JPG/PNG; AWS Textract OCR plus LLM extraction of seller, GSTIN, dates, taxes, totals and line items, processed asynchronously through a Redis job queue.
  • Automatic verification — live GSTIN check against the registry, arithmetic cross-check of subtotal + tax = total, required-field validation, duplicate detection and an explainable risk score.
  • Human-in-the-loop review — low-confidence or failed checks route the invoice to Needs Review with plain language reasons; reviewers correct fields, then Approve / Reject — every decision is audit-logged.
  • Ask your invoices (RAG) — natural-language questions answered from the user’s own invoices (Qdrant vector search), plus voice queries and invoice summarise & compare.
  • Tally export & analytics — Tally-ready CSV of purchase vouchers with export history; personal and company spend analytics with review-priority breakdowns.
  • SaaS foundations — multi-tenant RBAC, self-service signup with approval, usage metering, quota enforcement (HTTP 402), Razorpay subscriptions with signed webhooks, and a separate Super Admin panel for platform operations.

ENGINEERING HIGHLIGHTS

  • Correct by construction — usage metering and webhooks are idempotent through database UNIQUE constraints, so retries or duplicate events can never double-count or double-apply.
  • Webhook as source of truth — a paid plan activates only on a signed Razorpay webhook verified against the raw request body — never on a browser callback.
  • Privacy & security — strict tenant isolation, session-scoped unlock for password-protected uploads, scrubbed error tracking (no invoice contents, GSTINs or tokens leave the server).
  • Explainable AI — OCR confidence is recorded per figure; anything read below the threshold is flagged on screen with the exact reason instead of being silently trusted.
  • Production discipline — additive, auto-applied migrations; feature flags with safe defaults; 1,500+ automated backend tests; one-click deploys that roll back on a failed health check.

OUTCOME

InvoiceX runs in production on Oracle Cloud behind HTTPS with automated, health-gated deployments. It takes an invoice from upload to a reviewed, Tally-ready record, and its billing (free trial, metered quota, subscriptions) and Super Admin panel make it a complete, self-serve SaaS rather than a demo.