TaxCopilot is an AI-assisted platform for analysing Indian tax notices and preparing legal drafts. I worked across the frontend, backend, API gateway, data layer, and AI workflow.

The system is split into four services: a Next.js client, Node.js and Prisma backend with JWT authentication, a rate-limited API gateway, and a separate AI service.

AWS Textract extracts text from scanned documents, while PostgreSQL and pgvector retrieve relevant passages from Indian tax statutes. After generation, a citation-validation layer checks every referenced legal section against the retrieved material. An unverifiable citation triggers a warning and an automatic retry instead of silently returning an unsupported answer.

The project qualified for the prototype round of the AWS AI for Bharat Hackathon.

- [Case Study](https://tanveer-singh.vercel.app/work/taxcopilot)

- [GitHub](https://github.com/orgs/TaxCopilot/repositories)