-
-
Payment History Page
-
All orders placed screen
-
PayPilot AI assistant to help buy things easily
-
Product comparison page
-
Paypal Checkout Page
-
AI purchase Advisor
-
Initial Welcome Dashboard
Inspiration
Modern online shopping is broken by information overload. When shopping for electronics and tech gear, buyers often juggle dozens of open browser tabs, decipher convoluted spec sheets, cross-reference conflicting reviews, and second-guess whether they're actually getting a fair deal. We wanted to eliminate this "choice paralysis" by creating an intelligent, conversational agent that acts as an unbiased shopping co-pilot—guiding buyers from initial curiosity and deep technical comparison directly to a frictionless checkout.
What it does
PayPilot AI is an intelligent e-commerce and fintech companion that streamlines product discovery, decision-making, and purchasing:
Conversational AI Shopping Assistant: Answers natural-language questions regarding device specs, performance differences, budget recommendations, and use cases. Side-by-Side Product Comparison Matrix: Allows users to compare up to 4 devices simultaneously across key metrics like battery life, chipsets, display technology, and price-to-performance ratings. AI Purchase Advisor: Proactively analyzes discounts, highlights price drops, and surfaces actionable buying tips. Streamlined PayPal Checkout Flow: Bridges the discovery phase directly to payment preparation, generating simulated PayPal orders, calculating taxes, and logging transactions. Orders & Payment Ledger: Tracks delivery statuses and verified transaction histories in real time.
How we built it
We built PayPilot AI using a fast, decoupled full-stack architecture:
Backend & APIs: Built with Python and FastAPI, providing high-performance asynchronous REST endpoints for catalog queries, comparison generation, order handling, and assistant chat. Database: SQLite serves as our lightweight, embedded datastore managing the structured catalog, order states, payment receipts, and conversation history. AI Reasoning: Integrated Google Gemini 1.5 for context-aware conversational product advice, paired with a robust local heuristic fallback engine to ensure zero-downtime offline querying when an API key is not present. Frontend & Visual Experience: Developed with vanilla HTML5, modern CSS3, and JavaScript. We engineered an executive dark navy theme dashboard and an interactive 3D particle canvas splash screen on the landing page for an engaging visual entrance.
Challenges we ran into
Context Grounding vs. Hallucinations: Preventing LLM hallucinations regarding product pricing, availability, and hardware specifications was a major hurdle. We solved this by designing a retrieval-augmented prompting pattern that feeds live catalog data directly into the assistant's prompt context. Reliable Offline Fallback: We needed the app to stay fully functional even when external AI API keys or quotas weren't available. Building a deterministic, rule-based catalog matching engine that parses natural language queries and extracts intent locally required extensive testing. Responsive Multi-Item Matrix: Designing a dynamic comparison matrix that seamlessly scales from 1 to 4 complex products without breaking table alignments or overwhelming smaller screens took careful CSS grid and flexbox tuning.
Accomplishments that we're proud of
Seamless End-to-End Pipeline: Bridged the gap between conversational exploration and transactional fintech into one smooth user journey. Zero-Dependency Resilience: Created a dual-engine AI architecture that automatically pivots between cloud LLMs (Gemini) and the built-in local shopping engine without user-facing errors. Performance & Simplicity: Built a lightweight, snappy UI without heavy JavaScript frameworks, keeping initial load times near-instant. Immersive Visuals: Crafted a bespoke 3D particle animation for the landing experience that sets a futuristic, polished tone.
What we learned
Agentic Commerce Architecture: We learned how to structure structured catalog data and dynamic inventory feeds so LLMs can reason over them accurately in real time. State & Schema Design: Designing relational schemas in SQLite to handle both flexible product specifications and structured financial audit logs taught us clean schema normalization practices. Decoupled Frontend-Backend Synergy: Keeping the frontend pure and unopinionated made testing, mock payment generation, and API iteration significantly faster.
What's next for PayPilot
Live PayPal SDK & Multi-Gateway Integration: Transition from simulated sandbox checkout flows to live PayPal Smart Buttons, Apple Pay, and crypto payment rails. Real-Time Price Scraping & Tracking: Connect to live e-commerce APIs (Amazon, Best Buy, eBay) with automated price drop alerts sent via email or push notifications. Multi-Turn Cart Management: Enable the AI assistant to autonomously add items, apply coupon codes, and bundle compatible accessories directly via chat commands. Voice-Activated Shopping: Introduce real-time voice input and speech synthesis so users can shop and compare hands-free.
