Nova Mart

An AI e-commerce platform powered by Amazon Nova. It features an intelligent voice concierge, personalized smart recommendations, and a proactive AI voice agent to recover abandoned carts.

Comment

Inspiration

Modern e-commerce platforms are often static, transactional, and overly reliant on users to do the heavy lifting of finding exactly what they want. We realized that online shopping lacks the high-touch, hyper-personalized experience of an in-store personal shopper or concierge. We were deeply inspired by the recent advancements in conversational AI and the immense capabilities of the Amazon Nova model family. Our vision with Nova Mart was to bridge the gap between digital convenience and human-like customer service by introducing proactive, voice-first AI agents into every step of the retail journey.

What it does

Nova Mart is a next-generation, AI-driven e-commerce platform built around a sophisticated multi-agent architecture. Instead of just browsing a catalog, users can converse with our platform.

It features three primary AI capabilities:

  1. Interactive Voice Concierge: A sleek "Talk to Nova" interface allows users to search for products, ask for styling advice, or inquire about specifications using natural, conversational voice commands.
  2. Dynamic Recommendation Agent: An intelligent backend agent that analyzes user interactions in real-time to curate highly personalized, context-aware product feeds.
  3. Proactive Abandoned Cart AI: Our standout feature. If a user leaves items in their cart for a prolonged period, our specialized cart-recovery agent simulates a proactive, human-like voice call to the user to gently remind them, offer assistance, or provide a personalized discount to close the sale.

How we built it

We architected Nova Mart with a modern frontend framework deployed seamlessly on Vercel. The core intelligence is powered by Amazon Nova via Amazon Bedrock.

To handle the complexity of different tasks, we implemented a Master Agent pattern. The Master Agent parses user intent and routes it to specialized sub-agents: the Recommendation Agent to query data, or the Voice Search Agent to handle audio interactions. We utilized the Web Audio API to handle the microphone input and audio playback for the simulated voice features, incorporating subtle UI sound effects for a premium feel. The frontend was meticulously styled with a futuristic, sarvamAI-inspired aesthetic, utilizing vibrant cyan accents and fluid gradient backgrounds to make the interaction feel truly next-gen.

Challenges we ran into

Building a robust agentic workflow is incredibly challenging, and we faced several hurdles:

  • Multi-Agent Orchestration: Ensuring that the Master Agent seamlessly handed off context to the Abandoned Cart and Recommendation sub-agents without hallucinating or dropping user data required extensive prompt engineering and state management.
  • Audio & Voice Integration: We initially struggled with API key mismatches and managing the browser's Web Audio API permissions to ensure the voice overlay triggered flawlessly across different devices.
  • Deployment Hurdles: During the migration of our legacy codebase (formerly Stond Emporium) to the new Nova Mart brand, we ran into severe Vercel deployment issues, specifically involving Jekyll dependency conflicts, Liquid template exceptions, and bundler version mismatches. Debugging these build pipelines under a tight deadline tested our patience but ultimately strengthened our CI/CD knowledge.

Accomplishments that we're proud of

We are incredibly proud of the Proactive Abandoned Cart Agent. Seeing (and hearing!) the AI autonomously initiate a simulated phone call to recover a cart feels like a massive leap forward in e-commerce technology. We're also proud of how polished our UI turned out; resolving the complex routing and component state for the voice search overlay while maintaining a gorgeous 60fps responsive design was a huge win for the team.

What we learned

This project was a masterclass in AI integration. We learned how to deeply integrate Amazon Nova and manage complex Bedrock API interactions. We gained a solid understanding of how to implement the Master/Sub-agent architectural pattern. On the frontend, we sharpened our skills in Web Audio manipulation, state management for continuous microphone streams, and resolving complex Vercel build environment errors.

What's next for Nova Mart

We believe Nova Mart is just scratching the surface of proactive commerce. Our immediate next steps include:

  • Complex Support Handling: Expanding the agent's capabilities to autonomously handle returns, refunds, and order tracking entirely via voice.
  • Real-time Inventory Sync: Deeply integrating the Recommendation Agent with live supplier databases so it can suggest real-time alternatives if an item is out of stock.
  • Voice Personalization: Utilizing Amazon Nova's capabilities to adapt the AI's speaking style, language, and tone based on user demographics and past interactions.
  • Mobile Native App: Bringing the voice concierge to iOS and Android for a more integrated, hardware-level voice experience.

Built With

Share this project:

Updates