Inspiration
Online shopping often requires users to search through product catalogs, compare options, consider their budgets, and navigate checkout as separate steps. I wanted to explore how AI could make product discovery more natural while connecting the shopping journey to PayPal's payment infrastructure.
As the founder of LXNORO OJSC, I built PayPal LXNORO AI Agent to demonstrate a practical integration of conversational AI, product discovery, and PayPal checkout in one application.
What it does
PayPal LXNORO AI Agent is an AI-powered shopping assistant that helps users express their shopping needs in natural language, discover and filter catalog products, consider budget preferences, select a product, and initiate checkout through PayPal Sandbox.
Its main capabilities include:
- Conversational query enhancement: Uses Google Gemini Flash-Lite to optimize natural-language shopping requests.
- Product discovery: Connects to the Channel3 API to search and filter catalog results.
- PayPal checkout: Creates orders through PayPal Sandbox Orders v2 and verifies order status server-side before capture.
- Error diagnostics: Records structured PayPal API error details to support troubleshooting.
How we built it
I built the application using Python 3.11 and Streamlit. Google Gemini Flash-Lite handles shopping-query enhancement, while the Channel3 API provides product catalog access. The payment workflow integrates with PayPal's REST API using OAuth 2.0 authentication.
The backend includes input validation, decimal-based monetary calculations, approval URL domain validation, server-side order-status checks, and order capture handling. I also added a GitHub Actions workflow to validate Python syntax and published the source code under the MIT License.
Challenges we ran into
One of the main challenges was coordinating the checkout lifecycle across order creation, buyer approval, order-status verification, and capture. A browser redirect alone is not sufficient proof that a payment completed, so the integration checks the order state on the server before attempting capture.
Another challenge was validating the checkout workflow in a test environment. PayPal Sandbox transactions were completed and verified through PayPal Sandbox Activity. The current MVP returns to the order form after checkout rather than displaying a dedicated in-app success confirmation.
Accomplishments that we're proud of
- Built a working MVP that connects AI-assisted shopping queries, catalog discovery, and a PayPal Sandbox checkout workflow.
- Integrated Gemini, Channel3, and PayPal APIs into a single Streamlit application.
- Implemented server-side payment workflow safeguards, including order-status verification before capture.
- Verified test transactions through PayPal Sandbox Activity.
- Published the project source code with an MIT License and added automated Python syntax validation through GitHub Actions.
What we learned
This project reinforced the importance of separating AI-driven product discovery from payment authorization and verification. We learned that payment workflows must rely on server-side order state rather than treating a successful browser redirect as proof of payment.
We also gained practical experience integrating multiple APIs, handling external-service errors, validating monetary values, and documenting the limitations of an MVP clearly.
What's next for PayPal LXNORO AI Agent
The next improvements will focus on making the post-checkout experience clearer, adding more automated integration coverage, and improving product comparison and recommendation workflows. Further development can build on the current foundation while keeping payment verification server-side and preserving a clear separation between Sandbox testing and any future live-payment deployment.
Creator: NOORA AL-MAOLAI
Organization: LXNORO OJSC
Log in or sign up for Devpost to join the conversation.