Problem Statement

The hardest part of GST compliance is not filling in the numbers — it is preparing the right documents in the right format, consistently, every month, for every client. Most firms handle Statement 3 and Annexure B manually. Data is extracted from invoices, re-entered into Excel, mapped to the prescribed format, and uploaded to the portal. This process takes three to four hours per client and repeats every month without adding professional value.

The problem is not limited to time. A single format error, an invalid GSTIN, or an incorrect ITC classification results in a portal rejection or, worse, a GST notice. Rule-based tools classify well-structured entries correctly but silently pass through bad data. They inherit whatever the accountant entered and never interrogate whether the underlying document supports it.

No tool existed to handle the said work flow in a proficient and time saving manner.

Yet the obvious remedy — handing everything to an AI — is not acceptable in a professional context. A Chartered Accountant must be able to review and stand behind every figure that goes into a return or a refund claim. The challenge is therefore twofold: automate the mechanical work, and do so in a way that keeps the professional's judgment central to the output.

The Solution

The tool addresses this through three focused AI-powered modules — a Statement 3 Generator, an Annexure B Generator, and a GSTIN Validator — each built to automate a specific, high-frequency compliance task from end to end.

Statement 3 Generator

The user uploads raw purchase and sales invoice data in PDF or Excel format. Llama OCR extracts the relevant fields — GSTIN, invoice number, date, taxable value, tax amounts, and HSN codes — directly from the documents. Claude Code then validates the extracted data, catches inconsistencies, and maps each field to the correct column in the prescribed Statement 3 format using a model trained specifically on GST templates. The output is a portal-ready file. Exceptions and mismatches are flagged for CA review rather than silently passed through.

Annexure B Generator

The module accepts GSTR-2B data and cross-references it with the purchase register. Claude Code classifies each invoice by ITC eligibility, type of inward supply, and GSTR-2B period. A trained template engine generates a fully portal-compliant Excel file aligned with GSTN's 2026 mandate, ready for direct upload with Form RFD-01. Pre-submission checks catch duplicate entries, GSTIN format errors, date mismatches, and RCM classification issues before the file is produced — so the practitioner reviews clean output, not errors.

GSTIN Validator

The user submits an entire vendor master list. The tool verifies each GSTIN against the live GSTN API — confirming active registration status, validating the 15-digit format and embedded PAN, matching the registered name against firm records, and flagging state code mismatches. The output is a clean Excel report separating valid, invalid, cancelled, and mismatched GSTINs. Catching a cancelled GSTIN before ITC is claimed is materially less expensive than defending a disallowance after a notice.

Key Features

1. OCR-based document parsing. Llama OCR extracts invoice fields directly from PDFs and scanned documents. No manual data entry. No copy-paste.

2. AI validation at every step. Claude Code checks for GSTIN errors, field mismatches, ITC classification issues, and calculation inconsistencies before any output is generated.

3. Portal-compliant Excel for Annexure B. The trained template engine produces a Excel file that meets GSTN's 2026 mandate.

4. Bulk GSTIN verification. An entire vendor master is validated against the live GSTN API in a single run, with a clean report distinguishing valid, invalid, cancelled, and mismatched registrations.

5. Exceptions flagged, not hidden. Every output surfaces the entries that need professional review. The CA sees what the tool is uncertain about and retains judgment over the final submission.

6. Templates trained on GST formats. AI models trained specifically on Statement 3 and Annexure B structures ensure that field mapping and output format are correct regardless of variations in input data.

Technology Used

Relevance & Impact for the Profession

Every CA firm handling GST refund clients faces the same constraint: the time required to prepare Statement 3 and Annexure B manually scales with the number of clients, not with the complexity of the work. There is no professional judgment involved in re-entering invoice data into a template. There is real professional judgment involved in reviewing flagged exceptions, assessing ITC eligibility in edge cases, and deciding whether to claim a credit. The tool is designed to do the former so the practitioner can focus on the latter.

The GSTIN Validator directly addresses one of the most common triggers for GST notices — ITC claimed against a cancelled or invalid supplier registration. Validating the vendor master before filing takes minutes. Responding to a notice and defending the disallowance does not.

The governance model is intentional. The AI processes and flags; the Chartered Accountant reviews and approves. Every output is presented for professional review before submission. This keeps the practitioner accountable and in control — which is the only appropriate model for work that carries a professional signature.

These three tools are the first step. We have now started building more such AI-powered tools for GST, audit, and compliance workflows under SuvidhaAI — a platform being built specifically for the needs of CA firms and tax professionals. The goal remains the same: automate the mechanical work so practitioners can focus on the judgment that actually requires them.