-
-
CartPilot Homepage
-
CartPilot Homepage : Finds for TV
-
Product Details : Shows actual product information and available actions.
-
Product Comparison : Highlight the structured comparison experience.
-
AI Shopping Workspace : Intelligent Recommendations , Show ranked products, match scores, reasons, and trade-offs.
-
PayPal Login : Securely connecting to PayPal Sandbox to authorize the purchase.
-
Complete Your Purchase : Reviewing the order and approving the payment through PayPal Sandbox.
-
Order History : Shows a completed demo transaction
-
Order Confirmation : Successfully captured and verified payment, with the completed transaction reflected in CartPilot's order history.
-
Smart Cart : Show selected products, quantities, and totals.
-
Checkout Cancellation : Buyers remain in control. Cancelling the PayPal checkout returns them to CartPilot without completing the purchase.
-
Mobile Homepage - Shopping on the Go
-
Explore product information, pricing, and shopping actions through a responsive mobile interface.
-
Mobile AI Workspace - Your Shopping Copilot, Anywhere
Inspiration
Online shopping has become increasingly overwhelming. Finding the right product often means jumping between multiple websites, applying filters, comparing specifications, checking prices, and repeatedly adjusting search queries.
We asked ourselves: What if shopping worked more like having a conversation with someone who understands exactly what you need?
Instead of making users adapt to complicated search filters, we wanted to build an experience where they could simply describe what they were looking for, specify their budget and preferences, and let AI handle the discovery and comparison process.
That idea led to CartPilot — an AI-powered shopping companion designed to make online shopping more intuitive, personalized, and efficient.
What it does
CartPilot transforms natural-language shopping requests into an intelligent, guided shopping experience.
Users can describe what they need in their own words, including requirements, preferences, and budget. CartPilot interprets the request, searches live product listings, evaluates relevant options, and presents personalized recommendations with clear explanations.
Key features include:
- Conversational AI Shopping: Describe what you want instead of navigating endless filters.
- Intelligent Product Discovery: Search live product listings through Channel3.
- Context-Aware Recommendations: Evaluate products against budget, required features, preferences, and available product information.
- Transparent Comparisons: Compare shortlisted products and understand their trade-offs.
- Smart Cart Management: Add recommendations to your cart and manage quantities.
- PayPal Checkout: Complete a buyer-approved checkout through PayPal Sandbox.
- Order History: View completed demo transactions through locally stored order history.
Our goal is to make product discovery feel less like searching through a catalogue and more like having a personal shopping copilot.
How we built it
We built CartPilot using a modern web stack, combining AI orchestration, live product discovery, deterministic evaluation, and payment integration.
Technology stack:
- Frontend: Next.js, React, TypeScript, Tailwind CSS
- AI orchestration: Groq API with tool calling
- Product discovery: Channel3 SDK
- Payments: PayPal REST API (Sandbox)
- Data and persistence: Browser local storage for cart, comparison selections, and order history
How the system works:
- The user submits a natural-language shopping request along with optional budget and preferences.
- The AI interprets the request and determines the information needed for product discovery.
- The agent uses tool calling to interact with Channel3 and retrieve relevant product listings and details.
- A deterministic evaluation layer scores products against the user's requirements, budget, preferences, category relevance, and data completeness.
- CartPilot presents ranked recommendations with evidence-based explanations and trade-offs.
- Users can compare products, add their choices to the cart, and proceed to PayPal Sandbox checkout.
- Payment capture is performed server-side only after buyer approval and verification.
We deliberately separated AI-based intent interpretation from deterministic product evaluation. This helps keep recommendations grounded in retrieved product information rather than relying entirely on generated responses.
PayPal credentials and payment operations remain server-side, and the buyer retains control over approving the transaction.
Challenges we ran into
Building an agentic shopping experience involved challenges beyond simply connecting an AI model to a product API.
1. Making AI recommendations reliable
An AI response can sound convincing without being accurate. We needed recommendations to reflect actual product information, not invented specifications or prices. We addressed this by grounding recommendations in Channel3 results and using deterministic evaluation for ranking.
2. Understanding natural-language budgets
Users can express budgets in different ways, and explicit instructions may conflict with selected interface controls. We implemented consistent intent normalization and budget precedence so the same requirements are used throughout the discovery and evaluation process.
3. Integrating multiple services
Connecting AI tool calling, live product discovery, product details, cart management, and PayPal checkout required careful coordination between frontend and backend components. We also had to handle API failures, timeouts, missing credentials, and incomplete product information.
4. Making checkout secure and predictable
Payment integration required more than redirecting users to a payment page. We implemented server-side order creation, buyer approval, capture verification, and checks for the expected amount and currency. We also ensured that a cancelled or unsuccessful payment does not clear the cart.
5. Working within real-world API limitations
Currency and marketplace availability introduced constraints. Our current demonstration uses Channel3's supported US/USD product data and PayPal Sandbox rather than claiming international marketplace or live retailer fulfillment support.
Accomplishments that we're proud of
- Built a working agentic shopping journey from natural-language intent to product discovery, comparison, cart, and buyer-approved checkout.
- Integrated Channel3 for live product discovery instead of depending exclusively on static product data.
- Implemented AI tool calling to connect user intent with actual product-search operations.
- Combined AI interpretation with deterministic product scoring and transparent recommendation explanations.
- Integrated PayPal Sandbox with server-side payment verification.
- Built a responsive shopping interface with dedicated discovery, AI workspace, comparison, cart, checkout, and orders experiences.
- Added validation and error-handling mechanisms to make the experience more reliable.
- Achieved a passing automated test suite, typecheck, lint, and production build.
Most importantly, we are proud that CartPilot is not just an AI chatbot suggesting products. It connects conversational intent to real product discovery and a structured commerce workflow.
What we learned
This project taught us that building useful AI agents requires much more than prompt engineering.
We learned how to:
- Design tool-calling workflows that allow AI systems to interact with external services.
- Ground AI-generated recommendations in retrieved data.
- Combine probabilistic AI interpretation with deterministic business logic.
- Integrate third-party APIs while handling failures and service limitations.
- Design secure payment workflows with server-side validation.
- Build and test a complete full-stack application under hackathon time constraints.
We also learned that a good agentic experience should not remove user control. AI should reduce the effort required to make decisions, while users remain in control of their purchases and payments.
What's next for CartPilot
We see CartPilot evolving into a more capable personal commerce agent.
Our next steps include:
- Multi-marketplace discovery: Expand product discovery across more marketplaces and regions.
- Smarter personalization: Learn from user preferences, previous choices, and shopping patterns with appropriate user consent.
- Price intelligence: Add price tracking, historical trends, and deal discovery.
- Advanced agent workflows: Support multi-step shopping tasks, such as finding alternatives, checking availability, and refining recommendations through conversation.
- Persistent user accounts: Introduce secure accounts and server-side order history across devices.
- Broader payment support: Explore additional currencies and payment options as marketplace and provider support allows.
Our long-term vision is simple:
Shopping should begin with what you need, not with how well you can search for it.
CartPilot is our step toward that future.
Built With
- agentic-ai
- agentic-commerce
- ai
- ai-agents
- api-integration
- channel3
- data-driven-recommendations
- e-commerce
- full-stack
- groq
- llm
- natural-language-processing
- next.js
- paypal
- paypal-sandbox
- personalized-shopping
- product-discovery
- python
- react
- rest-api
- tailwind-css
- tool-calling
- typescript
- vite