Gokkul.dev
Back to all projects
Personal Project / Full Stack + AI · 2026

DIT

Document Injection Tool · AI document assistant

dit — experience preview (simulation)

Problem

Bills, invoices and letters arrive as photos, PDFs and email attachments, often with Telugu and English on the same page, and the due dates inside them get missed.

Approach

  • Bilingual OCR service (Python FastAPI + PaddleOCR) that runs Telugu and English recognisers and merges overlapping lines, because no single joint model reads both scripts.
  • Async processing pipeline on BullMQ + Redis (3 attempts, exponential backoff): OCR → Groq classification → field extraction. A failed OCR or AI call finishes the document with partial data instead of leaving it stuck.
  • Postgres + Prisma model of 25 entities (documents, bills, invoices, payments, reminders, tasks, insights, knowledge-graph relationships) with an intent-routed chat that cites its sources.
  • Gmail sync that pulls attachments, scheduled insight sweeps (missing recurring bills, expiring deadlines, price trends), Google Sign-In with AES-256-GCM token encryption, and 48 Vitest test files.

Result

Snap or forward a document and DIT reads it, files it, sets the reminder, and answers questions about it with a citation back to the source page.

Stack

Expo (React Native)TypeScriptNode.jsExpress.jsPostgreSQLPrismaBullMQRedisPython (FastAPI)PaddleOCRGroqGeminiGoogle OAuthVitest

The preview above replays the hardest moment of this project's Experience, drawn from the same simulation code. Click it to try the whole thing yourself.

Under the hood

A simplified excerpt of the core logic and how a request flows through the system.

More projects