💡 Inspiration Traditional e-commerce is full of friction. Users have to navigate complex menus, use rigid search bars, and click through multiple pages just to find a product and buy it. We realized that AI could make shopping as easy as texting a knowledgeable friend. However, conversational AI alone isn't enough—it needs the power to execute transactions. We built Aura to bridge the gap between intelligent product discovery and secure, frictionless payment execution.
⚙️ What it does Aura is an end-to-end agentic commerce bot. Users interact with the assistant via a natural chat interface to ask questions, compare tech gadgets, and narrow down their choices.
The moment a user decides to buy, the experience shifts from conversational to transactional. Aura autonomously extracts the purchase intent, item details, and pricing, and instantly generates a secure PayPal checkout link directly within the chat stream. The user simply clicks the link, authenticates via PayPal, and completes the purchase—no shopping carts or checkout forms required.
🛠️ How we built it Aura relies on a modern web stack designed for speed and reliability:
The Brain (Google Gemini API): We utilized Gemini via Google AI Studio as the core reasoning engine. Through strict system instructions, Gemini naturally converses with the user until it detects a concrete purchase decision. At that exact moment, it outputs a strict JSON payload containing the order details.
The Muscle (PayPal REST API): Our Node.js/Express backend listens for this specific JSON trigger. When detected, the backend autonomously authenticates with the PayPal Sandbox via OAuth2 and hits the /v2/checkout/orders endpoint, transforming the AI's data into a secure approve URL.
The Interface: The frontend is a fast, clean interface hosted on Vercel. It seamlessly parses standard text responses for conversation and dynamically renders clickable payment buttons when a transaction is ready.
🚧 Challenges we ran into The biggest hurdle was getting a Generative AI model to transition flawlessly from free-flowing conversation into a deterministic API trigger. Early on, the AI would generate conversational text alongside the JSON payload, which broke our backend parser. We solved this by rigorously refining our prompt engineering and system instructions to strictly enforce a pure JSON output the moment a purchase was confirmed.
We also had to carefully orchestrate the asynchronous flow between the frontend, the Gemini API, and the PayPal API to ensure the UI didn't freeze while generating the checkout link.
🏆 Accomplishments that we're proud of We successfully built a true "agent." Aura doesn't just chat; it takes autonomous action by talking to a financial gateway on the user's behalf. We are particularly proud of how seamless the handoff is between Gemini's natural language processing and PayPal's robust payment infrastructure.
📚 What we learned We learned how to design### 💡 Inspiration Traditional e-commerce is full of friction. Users have to navigate complex menus, use rigid search bars, and click through multiple pages just to find a product and buy it. We realized that AI could make shopping as easy as texting a knowledgeable friend. However, conversational AI alone isn't enough—it needs the power to execute transactions autonomously. We built Aura to bridge the gap between intelligent product discovery and secure, frictionless payment execution, creating a true "Agentic Commerce" experience.
⚙️ What it does Aura is an end-to-end agentic commerce bot. Users interact with the assistant via a natural chat interface to ask questions, compare tech gadgets, and narrow down their choices.
The moment a user decides to buy, the experience shifts from conversational to transactional. Aura autonomously extracts the purchase intent, item details, and pricing, and instantly generates a secure PayPal checkout link directly within the chat stream. The user simply clicks the link, authenticates via PayPal, and completes the purchase—no shopping carts or checkout forms required.
🛠️ How we built it Aura relies on a modern web stack designed for speed and reliability:
The Brain (Google Gemini API): We utilized Gemini via Google AI Studio as the core reasoning engine. Through strict system instructions, Gemini naturally converses with the user until it detects a concrete purchase decision. At that exact moment, it outputs a strict JSON payload ({"intent": "purchase", "item": "...", "price": ...}).
The Muscle (PayPal REST API): Our backend orchestrator listens for this specific JSON trigger. When detected, the backend autonomously authenticates with the PayPal Sandbox and hits the v2/checkout/orders endpoint, transforming the AI's data into a secure approve URL.
The Interface & Hosting: A clean HTML/JS frontend provides an iMessage-style interface that seamlessly parses standard text responses and renders clickable payment buttons. The application is deployed live using Vercel for fast, global accessibility.
🚧 Challenges we ran into The biggest hurdle was getting a Generative AI model to transition flawlessly from free-flowing conversation into a deterministic API trigger. Early on, the AI would generate conversational text alongside the JSON payload, which broke our backend parser. We solved this by refining our prompt engineering and system instructions in Google AI Studio to strictly enforce a pure JSON output the exact moment a purchase is confirmed.
🏆 Accomplishments that we're proud of We successfully built a true autonomous agent. Aura doesn't just chat; it takes action by talking to a financial gateway on the user's behalf. We are particularly proud of how seamless the handoff is between Gemini's natural language processing and PayPal's robust payment infrastructure, all happening in milliseconds without user friction.
đź“š What we learned Prompt Engineering for APIs: We learned how to design deterministic triggers out of non-deterministic AI models by strictly formatting system instructions.
Payment Gateway Integration: We gained deep experience with the PayPal Developer platform, learning how to manage OAuth tokens and dynamically generate payment intents server-side.
Full-Stack Orchestration: We learned how to efficiently route data between a frontend chat UI, an AI model, and a third-party financial API.
🚀 What's next for Aura This prototype proves the concept of agentic commerce. Moving forward, we want to add:
Multi-item shopping carts: Allowing the AI to bundle multiple products into a single PayPal checkout payload.
Real-time Inventory API: Connecting the agent to live stock levels so it only recommends available products.
Subscriptions: Using PayPal's billing agreements to let the AI set up recurring orders for consumables.
Built With
- agentic-ai
- artificial-intelligence
- css3
- e-commerce
- express.js
- google-gemini
- html5
- javascript
- node.js
- paypal
- rest-api
- tailwind-css
- typescript
- vercel
- zapier
