Naxia.AI Retail AI Store

An intelligent, Gemini-powered AI retail agent transforming everyday online shopping into personalized customer experiences. Agent to agent Agent to service Agent to Merchant no human in the loop

Comment

Inspiration

Modern e-commerce search tools often struggle to interpret complex, high-intent queries like "find a lightweight jacket under $100 that pairs with these boots." We created Naxia.AI Retail AI Store to bridge the gap between physical in-store assistance and digital shopping. Inspired by Gemini’s multimodal intelligence, our goal was to build an AI digital sales associate that truly understands visual context, inventory constraints, and real-time customer intent.

How we built it

  • Google AI Studio: Rapidly prototyped system instructions, tuned temperature parameters, and validated structured output schemas using Gemini models.
  • Core Microservices: Built backend services to seamlessly integrate the Gemini API with retail product catalogs and inventory databases.
  • Smart Matching Model: Implemented a multi-factor scoring algorithm to rank relevant products based on visual similarity, user constraints, and budget fit: $$S_{\text{item}} = \alpha \cdot \text{Similarity}(q, p) + \beta \cdot \max\left(0, 1 - \frac{|\text{Price} - \text{Budget}|}{\text{Budget}}\right)$$

Challenges we ran into

  • Latency vs. Accuracy: Balancing deep multimodal context analysis (processing high-res product photos alongside text) with low latency for instant responses.
  • Schema Integrity: Guaranteeing strict adherence to existing inventory data to prevent model hallucinations like non-existent items, outdated stock levels, or broken links.

Accomplishments that we're proud of

  • Designed a responsive, conversational UI that converts vague visual or natural language inputs into interactive product recommendations within seconds.
  • Standardized reliable system prompts using Google AI Studio to ensure consistent, clean Markdown outputs across various edge cases.

What we learned

  • Prompt Engineering in AI Studio: Iterating with context caching and few-shot examples inside Google AI Studio significantly improved the stability of structured tool calling.
  • Multimodal Depth: Combining visual product features with structured database metadata dramatically outperforms traditional keyword vector searches.

What's next for Naxia.AI Retail AI Store

  • Introducing secure long-term memory capabilities to recall returning shopper preferences and sizing.
  • Expanding voice-activated shopping navigation and localized multi-language support.

Built With

Share this project:

Updates