Jini

Private Document Intelligence — a local-first RAG workspace for personal documents.

Turn policies, statements, agreements, invoices, tax records, and other life-admin files into searchable answers, reminders, and structured insights.

Live demo · Developer documentation · Portfolio

Why this project matters

Jini demonstrates an end-to-end document intelligence workflow without requiring a hosted vector database or a paid model. It combines multi-format ingestion, ranked retrieval, citations, structured extraction, authentication, persistence, security controls, tests, and containerized delivery in one TypeScript codebase.

Engineering highlights

• Local-first processing and SQLite persistence keep the default workflow private and self-contained.
• Retrieval works without an API key; Groq synthesis is an optional enhancement rather than a hard dependency.
• Upload validation, rate limiting, defensive headers, secure cookies, and origin controls are built into the API.
• Thirty unit tests cover extraction utilities, ranking, parsing, and answer generation.
• The responsive application and built-in documentation are deployed under a reverse-proxy subpath.

---

Quick Start

npm install
cp .env.example .env
npm run dev

Open http://localhost:5173/jini/. The API runs at http://localhost:8788; Vite proxies /api to it.

Test accounts

• Test user: test@jini.local / JiniTest123!
• Guest: click Try live demo for one-click access with sample data

---

Features

• Upload PDF, DOCX, XLSX, CSV, TXT, Markdown, and JSON files
• Extract text, categories, tags, dates, amounts, and summaries automatically
• Search with ranked retrieval and document citations
• Ask questions with local extractive RAG (no API key required)
• Synthesize answers via Groq (optional, configure in Settings or .env)
• Track reminders from expiry, due, warranty, and renewal dates
• Surface high-value payments, subscriptions, tax coverage, and category mix
• Works offline — documents and data stay on your machine

---

Production Deployment

Docker (recommended)

docker build -t jini .
docker run -d \
  -p 8788:8788 \
  -v jini-data:/app/data \
  -e GROQ_API_KEY=gsk_your_key_here \
  jini

Or with docker-compose.yml:

services:
  jini:
    build: .
    ports:
      - "8788:8788"
    volumes:
      - jini-data:/app/data
    environment:
      - GROQ_API_KEY=gsk_your_key_here
      - NODE_ENV=production
    healthcheck:
      test: ["CMD", "wget", "-qO-", "http://localhost:8788/api/health"]
      interval: 30s
      timeout: 5s
      retries: 3

volumes:
  jini-data:

EC2 (without Docker)

Requires Node.js 22+.

git clone https://github.com/0xMudit/Jini.git
cd Jini
npm install
cp .env.example .env
nano .env              # add GROQ_API_KEY, adjust PORT if needed
npm run build
npm start

For a persistent process:

npm install -g pm2
pm2 start npm --name jini -- start
pm2 save

Open port 8788 in the EC2 security group, then visit http://<EC2_PUBLIC_IP>:8788.

Environment variables

Variable	Default	Description
PORT	8788	HTTP server port
VITE_API_BASE	same origin	Optional browser API origin; leave empty when using the Vite/nginx proxy
NODE_ENV	production	Set development for pretty logging
LOG_LEVEL	info	Pino log level (debug, info, warn, error)
GROQ_API_KEY	—	Groq API key for LLM synthesis
GROQ_MODEL	llama-3.3-70b-versatile	Groq model name
COOKIE_SECURE	false	Set true behind HTTPS
TRUST_PROXY	false	Set true behind a reverse proxy
ALLOWED_ORIGINS	—	Comma-separated CORS origins

---

Testing

npm test              # run once
npm run test:watch    # watch mode

30 unit tests covering text extraction, chunking, categorization, date/amount parsing, RAG scoring, and extractive answers.

---

CI/CD

Push to main triggers GitHub Actions:

1. Lint → Test → Build
2. Docker image built and pushed to ghcr.io/0xMudit/Jini:latest

---

Project Structure

server/
  index.ts          Express app, routes, middleware
  auth.ts           SQLite users, scrypt hashing, cookie sessions
  aiConfig.ts       Runtime Groq configuration
  extractors.ts     PDF, DOCX, XLSX, CSV, text extraction
  ingest.ts         Document processing pipeline
  insights.ts       Dashboard aggregates
  llm.ts            Groq answer synthesis
  rag.ts            Retrieval, scoring, extractive answers
  security.ts       CSP headers, rate limiting
  storage.ts        SQLite document/reminder storage
  textUtils.ts      Tokenization, chunking, date/amount parsing
  textUtils.test.ts Unit tests for text utilities
  rag.test.ts       Unit tests for RAG pipeline
src/
  Jini.tsx          Main app component
  types.ts          Shared TypeScript types
  lib/api.ts        API client, error handling, formatters
  components/       ErrorBoundary, EmptyState, Metric, etc.
  views/            AuthScreen, HomeView, AssistantView, etc.
data/
  uploads/          Uploaded files (gitignored)
  jini.sqlite       SQLite database (gitignored)

---

Tech Stack

Layer	Technology
Frontend	React 19, TypeScript, Vite 8, Lucide icons
Backend	Express 5, TypeScript, tsx
Database	SQLite (Node node:sqlite, WAL mode)
AI	Groq API (OpenAI-compatible), local extractive RAG
Logging	Pino with pino-pretty (dev)
Testing	Vitest
CI	GitHub Actions
Container	Docker (multi-stage, Alpine)