Spark — Turning Product Research into Confident Decisions
The Inspiration
Choosing the right product online should be simple, but the reality is often overwhelming.
People have to navigate multiple websites, compare specifications, investigate prices, evaluate product quality, and determine whether a product is actually worth buying. Important information is scattered across different sources, and making a confident decision can take hours.
We imagined a different experience: what if users could simply describe what they need, and an intelligent system could research the market, compare real products, explain the trade-offs, and help them complete their purchase in one place?
That idea became Spark — Global Product Intelligence.
Our goal was not to build another shopping chatbot. We wanted to create an intelligent product research and decision-making platform that connects AI-powered reasoning with real product data and actual payment infrastructure.
How We Built It
We approached Spark as an integrated system rather than a single AI conversation.
1. Multi-Agent Product Intelligence
At the heart of Spark is an AI-powered research workflow using Google Gemini 3.1 Flash Lite.
Instead of relying on one general-purpose prompt, we designed specialized research agents with distinct responsibilities, including:
- Intent and Requirements Intelligence
- Global Product Discovery
- Product Identity and Catalog Intelligence
- Technical Product Research
- Quality and Reputation Intelligence
- Pricing and Value Intelligence
- Independent Verification and Decision Intelligence
These specialists contribute to a structured research process coordinated by an orchestration layer.
Spark then brings the findings together into its Omega Decision Report, covering product specifications, performance, advantages, disadvantages, pricing, evaluation, comparisons, recommendations, and supporting sources.
2. Real Product Data with Channel3
An important design decision was to connect AI reasoning to external product information.
We integrated Channel3 to support product discovery and catalog research. This helps Spark ground its analysis in actual product information rather than depending entirely on information generated by an AI model.
3. Payments with PayPal
We integrated the PayPal Developer Platform using its free Sandbox environment.
The integration uses PayPal Checkout and the Orders API v2 to support the purchase workflow. Users can select products, proceed to checkout, and explicitly approve test transactions.
We also used the APIMatic PayPal Context Plugin during development to help understand PayPal API documentation and implementation requirements.
4. User Accounts and Application Infrastructure
We used Firebase Authentication and Firestore to support user accounts and application data, including the foundation for personalized shopping experiences and purchase history.
The application is deployed on Vercel, making the Spark experience accessible through a live web application.
The Challenges We Faced
Building Spark involved challenges that went far beyond connecting an AI model to a user interface.
Reliable product intelligence: AI-generated information is not automatically accurate. We had to think carefully about product identity, source verification, pricing, availability, and the difference between confirmed information and missing evidence.
Multi-agent coordination: Dividing product research into specialized responsibilities introduced complexity. The agents needed to work with consistent requirements and pass useful findings between stages without losing the user's original intent.
Real payment integration: Connecting PayPal was more demanding than simply displaying a payment button. We had to understand the order lifecycle, Sandbox buyer approval, order capture, and the relationship between payment status and purchase records.
Application consistency: Authentication, user data, subscription plans, usage limits, and purchases all needed to work together. Keeping these systems consistent across the application was an important engineering challenge.
These challenges taught us that a convincing AI demonstration is very different from building a dependable AI-powered product.
What We Learned
The most important lesson was that AI intelligence needs reliable data, verification, and well-designed execution workflows.
We learned to think beyond prompts and model responses. Building a useful AI product requires system architecture, specialized agents, external integrations, error handling, data consistency, and extensive testing.
We also learned that payment integration is not complete just because an order can be created. The entire user journey—from selecting a product to approving a payment and recording the resulting transaction—must be considered.
Most importantly, we learned how to turn an ambitious idea into a connected product involving AI research, external data, application infrastructure, and payment technology.
What's Next
Spark is designed to grow beyond product search.
Our vision is to make it a global product intelligence platform where people can research, evaluate, compare, and purchase products with greater confidence.
We want every recommendation to be explainable, every important claim to be supported by evidence, and every purchase to remain under the user's control.
Spark is not just about finding products. It's about helping people make better decisions.
Built with curiosity, experimentation, and the ambition to make AI genuinely useful.
