SpiritSage

Architecture Diagram
Homepage with Gemini Live

SpiritSage: The AI-Powered Sommelier for the Modern Liquor Store Inspiration The journey began with the Google Antigravity platform, which gave me the confidence to bridge the gap between "tech enthusiast" and "solution builder." Inspired by the power of the Gemini ecosystem, I wanted to solve a high-volume retail challenge. My friend’s liquor store, with over 3,500 unique SKUs, was the perfect canvas. I saw an opportunity to turn a complex inventory into a conversational, "live" shopping assistant.

What it does SpiritSage is a multimodal AI agent that acts as both a digital salesman and a knowledgeable co-worker.

Live Interaction: Users engage with the agent via Gemini’s Audio/Video API for real-time recommendations.

Trend Discovery: It cross-references current TikTok beverage trends to suggest what’s "viral" right now.

Smart Alternatives: Using fuzzy matching, it provides intelligent substitutions if a specific brand is out of stock.

Personalized Concierge: It transforms a cold inventory list into a warm, personalized experience, helping customers find the perfect bottle through natural dialogue.

How we built it I leveraged the Antigravity agent builder and the Google Cloud ecosystem to create a production-ready stack without writing traditional code.

Infrastructure: Hosted entirely on Google Cloud Run for scalability.

AI Engine: Powered by Gemini 2.0 Flash via Vertex AI, utilizing the Multimodal Live API for low-latency voice and vision.

Grounding & Accuracy: To prevent hallucinations, I implemented Google Search Grounding for trends and strictly mapped product queries to a verified inventory.json dataset.

Architecture: A Node.js/Express backend handles the logic, while a React (Vite) frontend provides a sleek, responsive interface.

Challenges we ran into The primary challenge was prompt-based precision. Initially, the agent struggled to maintain strict inventory boundaries. I learned to "collaborate" with Gemini—using separate chat sessions to troubleshoot the UI build-out and refine the system instructions. This iterative process taught me how to guide an AI tool to deliver exactly what a business owner needs, even when the underlying logic is complex.

Accomplishments that we're proud of As a non-developer, my proudest achievement is building a functional, cloud-native enterprise tool from scratch. Partnering with AI allowed me to bypass the "coding barrier" and focus entirely on user experience and business logic. Seeing a complex 4,000-item recommendation engine run with sub-millisecond performance was a true "lightbulb" moment.

What we learned This project proved that AI is the ultimate equalizer. It turns "dreams into reality" by allowing anyone with a vision and the right tools (like Gemini and Antigravity) to build sophisticated technology that solves real-world problems.

What's next for SpiritSage The next step is a live pilot installment at my friend’s retail location. I plan to gather direct customer feedback to refine the "Salesman Persona" and explore how real-time vision can help with shelf-stocking and inventory audits.