Alan Abishek
← All work

Case 01

R.K. Tantry & Co. · Apr–Jun 2026

Cut GST filing from 4 days to 4 hours

  • 95% faster
  • 4d → 4h
  • Shipped

Engineered an AI-powered GST automation platform that collapsed manual tax filing from days of spreadsheet work into a same-day pipeline.

Led full-stack + AI orchestration

  • Python
  • FastAPI
  • Next.js
  • LLMs
  • n8n
  1. 01 Hook

    Tax teams were burning four full days every filing cycle on work a machine could own.

  2. 02 Context

    AI Agent Developer Intern at R.K. Tantry & Co. — building automation for financial and tax workflows with a small engineering surface and real filing deadlines.

  3. 03 Problem

    Manual GST preparation meant copying invoices, reconciling mismatches, and babysitting spreadsheets. Cycle time sat at ~4 days; error risk rose with every handoff.

  4. 04 Insight

    The bottleneck wasn't 'more AI' — it was a deterministic pipeline with LLM assist only where documents were messy, plus a clear action queue for humans.

  5. 05 Exploration

    Tried pure spreadsheet macros, then chat-only agents. Macros couldn't read messy invoices; chat agents hallucinated fields without a validation layer.

  6. 06 Decision

    Owned the trade-off: slower on creative edge cases, faster and safer on the 90% path — OCR/extract → validate → reconcile → human-only exceptions.

  7. 07 Solution

    Shipped an end-to-end GST automation platform: ingestion, validation, reconciliation, and a dashboard so operators finish filing in hours, not days.

  8. 08 Impact

    Filing time dropped from 4 days to 4 hours (~95% efficiency gain). The team stopped living in the spreadsheet and started reviewing exceptions only.

Impact

  • 95%

    Efficiency gain

  • 4h

    Filing time (was 4 days)

  • 1

    Unified pipeline

  • Shipped

    Production use