aura

🎤 Aura AI: The Future of Voice-First, Personalized, Agentic Shopping

Aura AI fixes what traditional shopping platforms never solved:
They overwhelm you with thousands of irrelevant results and never understand your taste, context, or how items actually look on you.

Aura replaces this broken experience with a voice-driven, multi-agent shopping assistant that reasons about your needs, learns your preferences, and even generates customized outfit visualizations on your own photo—something no e-commerce platform offers today.

🧠 Multi-Agent System Powered by LangGraph

Aura AI is built on a LangGraph-based multi-agent orchestration that gives structure, determinism, and deep context retention across every conversation. Each agent handles a specific domain:

🕵️ Research Agent

Pulls structured products from Google Shopping Search, filters noise, and finds items that fit your context—occasion, location, style, and budget.

🎨 Styling Agent

Takes the selected products and merges them onto the user’s own photo, using NanoBanana Pro image generation.
This delivers the single biggest differentiator:
Instead of guessing, the user actually sees the outfit on themselves.

📊 Preference / Ranking Agent

Learns user taste over time using embeddings.
Every choice, skip, like, and refinement updates a vector profile—meaning:
The more you shop, the better Aura gets at predicting what you’ll love and re-ranks future products accordingly.

This gives Aura an evolving sense of personal style that traditional platforms simply cannot build.

🙋‍♂️ Human-In-the-Loop (HITL) Refinement

Aura doesn’t hallucinate.
It actively asks clarifying questions to refine user intent:

“Is this for a trip or a wedding?”

“Should we keep it under ₹2000?”

“Do you prefer printed or solid colors?”

This HITL pattern makes the system accurate, trustworthy, and highly personalized.

🎤 Voice-First Frontend (No Voice Agent)

Voice is integrated client-side using ElevenLabs, enabling a fully hands-free, natural shopping flow.
The intelligence remains in the backend—the frontend simply becomes the smoothest entry point.

🚀 Production-Ready Engineering

Aura was built as a real product, not a demo:

Google Cloud Run for fast, horizontally scalable deployment

Fully automated CI/CD using GitHub Actions

Containerized ChatAPI + agent graph

Embedding service for preference learning

This gives Aura the performance and reliability required for a real e-commerce platform.

Since we were excited about this hackathon, we had already decided that we wanted to build a voice-based, agentic shopping platform. We knew we would need multiple agents, so before the event we experimented with LangGraph to validate multi-agent orchestration patterns.

During the hackathon, we built the Chat APIs, the frontend using Bolt, and the ElevenLabs voice integrations. We also developed the Research Agent to pull product data from the Google Shopping Search API, and the Styling Agent to merge searched products onto user photos using NanoBanana Pro. All of this work came together to form a fully multimodal shopping platform.