Document Assistant for a 40-Person Accounting Firm
Staff ask questions about GST circulars, direct tax rulings, and client audit files in plain English. Answers cite the exact PDF page and paragraph they originated from.
40 accountants and article clerks were spending an estimated 25 to 45 minutes manually searching hundreds of annual Indian taxation circulars, CBDT notifications, and past client audit reports before drafting advisory opinions.
Stackmate built a private, high-accuracy RAG pipeline using Claude 3.5 Sonnet, pgvector on PostgreSQL, and a Next.js internal portal. Documents are parsed with layout-aware chunking, and every generated response includes inline clickable citations linking to the exact source page.
“Our clerks now verify past tax rulings in seconds instead of half the afternoon. The citations give our senior partners the confidence to trust the output.”
— V. Ramaswamy, Managing Partner