# The Compliance Stack Nobody Talks About > Source: > Published: 2026-08-28 06:39:30+00:00 GST returns, TDS reconciliation, ROC filings , for most Indian startups this is still a spreadsheet problem. But a quiet shift is happening. Companies are now running AI agents that: Pull GSTR-2B data from the GST portal API nightly Match it against internal purchase ledgers Flag ITC mismatches before the human accountant logs in Monday morning Generate a pre-filled GSTR-3B with confidence scores on each line item The agent doesn't file. It prepares. The human reviews and approves. But the 3 hours of reconciliation work is now 15 minutes of exception handling. What's Actually Powering This Most implementations use a combination of: Python scripts calling GSTN sandbox APIs for return data Claude or GPT-4o for ledger interpretation and anomaly flagging Simple rule engines for TDS section classification (194C vs 194J vs 194Q) Email triggers for due-date reminders with pre-filled data None of this requires a dedicated ML team. A single backend developer with access to the GSTN API and an LLM SDK can build a working prototype in a weekend. The Interesting Problem: Section Classification The hardest part is TDS section classification. When a company pays a vendor, is it 194C (contractor) or 194J (professional/technical)? Misclassify, and the TDS rate changes (1-2% vs 10%). An LLM with the right context window can handle this with ~90% accuracy , the edge cases still need human review. Where This Is Going The CBDT has announced a move toward faceless compliance. Automated matching systems already flag mismatches before they become notices. The startups building internal compliance automation now will have a significant operational advantage in 3 years. Read more about how [AI in tax compliance India](https://taxgarden.in/blog/ai-in-tax-compliance-india-2026) is reshaping accounting workflows for SMEs and startups.