AI-Pass - AI Task Workspace

A full-stack web application for submitting text-based tasks, running AI-powered analysis, and viewing structured results.

---

Architecture

Stack

Frontend

• React 19 + TypeScript
• React Router DOM
• Plain CSS (no framework)
• Vite

Backend

• Node.js + Express + TypeScript
• Prisma ORM (v6.19.2)
• PostgreSQL (Neon - serverless)
• JWT authentication
• bcryptjs for password hashing

---

Database Design

users

Field	Type	Notes
id	UUID	Primary key
email	String	Unique
password	String	bcrypt hashed
createdAt	DateTime	Auto

tasks

Field	Type	Notes
id	UUID	Primary key
title	String
taskType	String	summarize / classify_risk / extract_info
inputText	String
status	String	pending / running / completed
createdAt	DateTime	Auto
userId	UUID	Foreign key → users

task_results

Field	Type	Notes
id	UUID	Primary key
resultText	String	Human-readable summary
resultJson	JSON	Structured output
createdAt	DateTime	Auto
taskId	UUID	Foreign key → tasks (unique)

---

API Endpoints

Method	Endpoint	Auth	Description
POST	/auth/register	No	Register new user
POST	/auth/login	No	Login, returns JWT
GET	/tasks	Yes	List all tasks for user
POST	/tasks	Yes	Create a new task
GET	/tasks/stats	Yes	Usage summary
GET	/tasks/:id	Yes	Task detail with result
POST	/tasks/:id/run	Yes	Execute AI analysis

---

AI Engine

The AI processing layer is implemented as a context-aware rule-based mock engine located in src/services/ai.ts.

It is intentionally architected as a drop-in replacement for any LLM API. Swapping to a real provider requires only replacing the contents of src/services/ai.ts - no changes to routes or database.

How it works

Summarize - splits input into sentences, extracts key points from actual content, returns word count and sentence count.

Classify risk - scans for a tiered keyword dictionary (critical / high / medium terms), returns PASS / FAIL / NEEDS_INFO decision with LOW / MEDIUM / HIGH / CRITICAL risk level and specific risk factors found in the text.

Extract info - uses regex to extract emails, URLs, phone numbers, dates, names, and reference numbers from raw text.

All responses include decision, confidence, and reasons fields matching the assignment's required structured output shape.

A simulated network delay of 800–1400ms is applied to mimic real API latency.

To plug in a real API

Replace src/services/ai.ts with any of the following - the function signature stays identical:

• OpenAI (gpt-4o-mini) - requires billing
• Anthropic Claude (claude-haiku-4-5-20251001) - requires credits
• Google Gemini (gemini-1.5-flash) - free tier, may have regional restrictions

---

Running Locally

Prerequisites

• Node.js v18+
• PostgreSQL database (or Neon free tier)

Backend

cd backend
npm install

Create .env from .env.example:

DATABASE_URL="your-postgres-connection-string"
JWT_SECRET="your-secret-key"

npx prisma@6 migrate dev
npm run dev

Server runs at http://localhost:4000

Frontend

cd frontend
npm install
npm run dev

App runs at http://localhost:5173

---

Environment Variables

Variable	Required	Description
DATABASE_URL	Yes	PostgreSQL connection string
JWT_SECRET	Yes	Secret for signing JWTs
PORT	No	Server port (default 4000)

---

What I Would Improve Next

• Plug in a real LLM API once network/billing constraints are resolved
• Add pagination to task history
• WebSocket support for real-time task status updates
• Rate limiting on auth endpoints
• Refresh token mechanism
• Export task results as PDF or CSV
• Role-based access for team workspaces
• Better UI

---

Live URLs

• Frontend: https://ai-pass-frontend-8s3x.onrender.com
• Backend: https://ai-pass-backend.onrender.com