Traditional tax compliance systems in India are built for recording transactions, not for resolving anomalies. Every month, corporate accounting teams must manually match GSTR-2B files—containing real-time uploads from vendors—with their internal purchase ledgers. Any mismatch, whether due to late vendor filings, incorrect GSTINs, or rounded-off totals, directly impacts the company's eligible Input Tax Credit (ITC), locking up working capital.
At ShipLabs, we designed a deterministic multi-agent pipeline to solve this. Instead of simple keyword matching or rigid spreadsheet formulas, the system utilizes a combination of advanced optical character recognition (OCR) and an agentic decision tree to perform deep semantic reconciliation.
The ingestion layer handles diverse, noisy data sources: scanned thermal paper receipts, mobile camera captures from transport partners, and standard PDF invoices. It extracts tabular data with 99.8% precision, converting unstructured layouts into standardized transaction schemas.
Once normalized, the matching engine runs an agentic reconciliation loop. It handles edge cases like split payments, misspelled vendor trade names, and date shifts (e.g., invoice dated April 30th but filed in GSTR-2B in May). If the system detects a mismatch, it doesn't just raise a flag—it initiates resolution. The system drafts a context-aware email to the specific vendor detailing the exact mismatch code, the transaction date, and the corrective action required.
By translating probabilistic language-model understanding into highly disciplined, deterministic business workflows, our clients are reclaiming locked capital without human intervention.