Skip to main content

Validated Accounting AI

Autonomous bookkeeping automation and deterministic statutory compliance for accounting practices.

High-volume accounting, audit, and tax advisory practices in the UAE face relentless manual backlogs across receipt capture, multi-currency VAT classification, and bank reconciliation under Federal Decree-Law No. 47 of 2022 (Corporate Tax) and Cabinet Decision No. 52 of 2017 (VAT Executive Regulations). Generic OCR and standard LLM prompts fail on 20–30% of real-world edge cases. Dotnitron engineered an autonomous accounting engine that automates 80%+ of repetitive voucher entry while enforcing strict human governance, integer-money arithmetic checksums, and audit defensibility.

Project status

Validated pipeline, reference architecture, and simulated benchmark suites

  • Deterministic FTA Rules Engine
  • Native ERP & WhatsApp Pipeline
  • 100% Arithmetic & TRN Defensibility

How we contributed

Dotnitron architected and developed the end-to-end accounting engine: multi-modal vision extraction, deterministic statutory rule enforcement (15-digit TRN syntax, Cabinet Decision No. 52 of 2017 Art. 59 full invoice verification, Corporate Tax Federal Decree-Law No. 47 Arts. 32 & 33 entertainment splits and penalty disallowances), integer-money arithmetic checksums (Delta <= 0.02 AED), WhatsApp Business receipt intake, native bidirectional sync for Zoho Books & Xero, and API connectors for QuickBooks Online & Tally Prime.

Challenge

Autonomous accounting and bookkeeping engine for multi-entity firms

Standard tools (Dext, AutoEntry, or naive LLM prompts) break on Gulf accounting realities: faded thermal receipts, bilingual Arabic/English invoices with Eastern Arabic numerals, split VAT rates on mixed invoices, missing buyer TRNs on bills exceeding AED 10,000 (Cabinet Decision No. 52 Art. 59), and non-standard bank feed narrations. An unsupervised 2% error rate exposes accounting practices to full input VAT disallowance under VAT Law Art. 55 and crippling Federal Tax Authority (FTA) audit penalties under Cabinet Decision No. 49 of 2021.

Validated Accounting AI

About this client work

We protect the customer's identity and confidential implementation details. This project had been built and validated, but had not yet reached its first end-client deployment.

What we built

From product design to a validated system.

Dotnitron delivered these parts of the product before its first end-client deployment.

Step 01

Engineered the Dual-Guard Pattern: an AI multimodal vision extractor for field perception paired with an immutable deterministic validation engine that blocks hallucinations before they touch the general ledger.

Step 02

Embedded UAE FTA statutory rules: automated 15-digit TRN syntax validation, Cabinet Decision No. 52 of 2017 Art. 59 full invoice enforcement (>AED 10,000 threshold), and Corporate Tax Decree-Law No. 47 of 2022 Arts. 32 & 33 automated 50/50 entertainment splits and penalty disallowances.

Step 03

Designed the zero-friction Gulf operational workflow: client receipt intake directly via WhatsApp Business API, routing into native Zoho Books & Xero 'Draft Bills' queues (with API connectors for QuickBooks Online & Tally Prime) with raw vouchers attached for 5-second senior CA sign-off.

Step 04

Built integer-money arithmetic checksums (Subtotal + Tax == Total with Delta <= 0.02 AED) and Central Bank of the UAE (CBUAE) daily exchange rate feeds (VAT Law Art. 69) for statutory foreign currency conversions.

What was validated

What the customer had tested and accepted.

The product was complete enough for customer validation and demonstrations. Its first end-client deployment was still pending.

Results and delivery milestones

These were the delivery and validation milestones completed.

  • Over 80% reduction in manual data entry and voucher typing across benchmark accounting workloads
  • 100% of arithmetic mismatches and invalid TRNs deterministically blocked before ERP draft creation; zero unverified calculations touch the general ledger
  • Tri-tier human governance workflow preserving 100% Chartered Accountant oversight and statutory audit defensibility
  • Turnaround time for client WhatsApp receipt processing reduced from 48-72 hours down to under 5 minutes into draft state

Inside the product

What the validated platform includes.

We describe the product without exposing confidential architecture, business information, or customer data.

Capability 01

Autonomous extraction and line-item categorization for complex bilingual Arabic/English tax invoices

Capability 02

Deterministic tax verification flagging high-value invoices (>AED 10k) missing buyer TRN to prevent FTA input-tax disallowance under Art. 59

Capability 03

Native ERP draft bills with original receipt evidence attached and automated 50/50 Corporate Tax Art. 32 splits applied

Scale and scope

The scale of the work.

These figures reflect specific points during our involvement. They should not be read as current usage or independently audited business outcomes.

Measure 01

>80% keystroke reduction

Measure 02

100% arithmetic & TRN verification

Measure 03

Zero unverified ledger writes

Important context

This case study and reference architecture reflect Dotnitron's engineering implementations, deterministic rule frameworks, and simulated benchmark suites across UAE tax regimes. Specific operational time savings and straight-through processing rates vary based on individual firm document quality, client receipt legibility, and existing ERP configuration.

  • Deterministic FTA Rules Engine
  • Native ERP & WhatsApp Pipeline
  • 100% Arithmetic & TRN Defensibility

Have something in mind?

Have a difficult AI product or system to build?

Tell us what is not working today, who needs a better experience, and what you want the business to be able to do.