Architecture of NovaTrack

NovaTrack — AI Price Intelligence

Inspiration

Growing up in India, I watched family members spend hours refreshing Amazon and Flipkart pages, waiting for a price to drop before a festival or a big purchase. The frustration was real — prices fluctuate daily, deals vanish in hours, and no one has time to manually monitor dozens of products across multiple platforms.

When I discovered Amazon Nova Act — an AI agent capable of controlling a real browser with natural language instructions — I immediately saw the solution. What if an AI could do that refreshing for you? Not with brittle CSS selectors that break every time a website redesigns, but with genuine visual intelligence — reading a page the way a human would?

That insight became NovaTrack: an AI price hunter built specifically for shoppers, powered by Amazon Nova.

What I Learned

This project taught me more than any tutorial could:

How AI agents work in practice — Nova Act doesn't scrape HTML. It understands pages visually, which means it handles JavaScript-heavy sites, dynamic popups, and layout changes gracefully. This completely changed how I think about web automation.
The real cost of fragility — Traditional scrapers break constantly. A single CSS class rename on Amazon's product page can kill your entire tracker. Vision-based AI eliminates this class of problem entirely.
Browser automation at scale — Managing Playwright sessions, handling timeouts, and dealing with bot detection taught me how much complexity hides behind "just visit a website."
Full-stack product thinking — Building end-to-end from a landing page to a live dashboard to email alerts forced me to think like a product designer, not just a developer.
How exchange rates and locale formatting matter — Displaying prices in ₹ Indian Rupees required understanding Intl.NumberFormat, locale-aware formatting, and live USD→INR conversion — small details that make a product feel native.

Mathematically, the savings calculation is straightforward:

$$\text{Savings} = P_{\text{initial}} - P_{\text{current}}$$

$$\text{Drop \%} = \left(\frac{P_{\text{initial}} - P_{\text{current}}}{P_{\text{initial}}}\right) \times 100$$

Where $P_{\text{initial}}$ is the price when tracking began, and $P_{\text{current}}$ is the latest extracted price.

How I Built It

NovaTrack is a full-stack web application with three distinct layers:

1. AI Agent Layer — Amazon Nova Act

The core intelligence. When a user submits a product URL, Nova Act:

Launches a real Chromium browser instance
Navigates to the URL with a realistic user agent
Receives a natural language instruction to find and return the current price as structured JSON
Returns { price, currency, title, imageUrl } — no selectors, no DOM parsing

const result = await client.act(
  'Find the current selling price. Return JSON: ' +
  '{"price": <number>, "currency": "USD", "title": "<name>"}'
);

2. Backend — Node.js + Express

REST API with endpoints for adding, deleting, and manually checking items
node-cron scheduler runs price checks every hour automatically
Nodemailer sends styled HTML alert emails when target prices are hit
JSON file storage — lightweight and portable for a hackathon prototype
dotenv for secure credential management

3. Frontend — React 18

Landing page with animated starfield background, gradient hero headline, How It Works section, and Features grid
Dashboard with live stats bar, product cards, price progress bars, and alert history
Sign In / Sign Up flow with Google and email/password options
All prices displayed in ₹ Indian Rupees using Intl.NumberFormat('en-IN')
Live AI Agent Log terminal at the bottom showing real-time browser activity

Architecture Diagram

User Browser
     │
     ▼
React Frontend (localhost:3000)
     │  REST API calls
     ▼
Express Backend (localhost:3001)
     │
     ├── POST /api/items ──► Nova Act Agent
     │                            │
     │                       Chromium Browser
     │                            │
     │                       Product Page
     │                            │
     │                       AI Price Extraction
     │                            │
     └◄───────────────────── Structured JSON
     │
     ├── node-cron (hourly) ──► Same flow above
     │
     └── Nodemailer ──► Email Alert to User

Challenges I Faced

1. Nova Act Node.js SDK availability

The biggest hurdle: Amazon Nova Act only has a Python SDK at the time of building. The Node.js package @aws/nova-act does not yet exist on npm. I solved this by replicating the exact same architecture using Playwright (real Chromium browser) + Groq-hosted Llama 3 (free AI inference) for the hackathon demo, with the backend designed as a drop-in module — swapping Nova Act in requires changing a single file.

2. Major retailers blocking headless browsers

Amazon, Best Buy, and Flipkart actively detect and block Playwright's default headless mode. I addressed this by:

Setting headless: false to run a visible browser
Injecting a script to mask the navigator.webdriver flag
Using a realistic userAgent string
Adding a 2–3 second settle delay after page load

3. AI model deprecations mid-build

Two AI models were deprecated during development:

llama3-8b-8192 on Groq → switched to llama-3.3-70b-versatile
gemini-1.5-flash → switched to gemini-2.0-flash

This reinforced the importance of abstracting the AI layer behind a service module.

4. Currency formatting for Indian users

Indian number formatting (lakhs, crores) is different from Western conventions. ₹1,00,000 is one lakh — not one hundred thousand. Using Intl.NumberFormat('en-IN', { style: 'currency', currency: 'INR' }) handled this correctly across all components.

5. Free-tier API quota limits

Gemini's free tier had limit: 0 on a fresh API key until quota was activated. This required switching providers mid-build and reinforced the value of provider-agnostic architecture.

What's Next

Python backend using the official Nova Act SDK for true browser AI
Mobile app with push notifications
Price history charts with trend analysis
Multi-product comparison — track the same product across Flipkart, Amazon, and Croma simultaneously
Browser extension — one-click tracking from any product page

Built with ❤️ for shoppers. #AmazonNova