NexusAI — Enterprise Agentic Fashion Assistant 🚀

A production-ready, intelligent retail assistant demonstrating the future of e-commerce. Powered by LangGraph agentic routing, advanced RAG (Retrieval-Augmented Generation) with semantic vector search, and Groq's lightning-fast inference.

🔴 Live Vercel Demo · Report a Bug

---

🌟 The Vision: Why NexusAI?

Modern e-commerce requires more than just keyword search—it demands intelligent, context-aware conversational agents. NexusAI bridges the gap between a personal shopper and a customer support representative.

Instead of relying on rigid decision trees, NexusAI utilizes an Agentic AI architecture capable of understanding user intent, performing semantic database searches, reranking results for maximum relevance, and streaming personalized responses in real-time.

🏗️ Enterprise Architecture & Tech Stack

NexusAI is built on a highly decoupled, modern microservices architecture optimized for speed, scalability, and zero-hallucination AI responses.

1. The Brain: Agentic AI Workflow (LangGraph & Groq)

• LangGraph Routing: Every incoming message passes through a deterministic router node. The agent autonomously classifies the intent into SALES, SUPPORT, or OUT_OF_SCOPE, ensuring the LLM only uses the exact tools and context required for that specific query.
• Groq Llama-3 Inference: Replaces traditional slow LLM endpoints with Groq's LPU (Language Processing Unit), achieving sub-second token generation for a fluid, real-time user experience.
• Conversation Memory: Context window injection ensures the agent remembers the last 5 conversation turns, allowing for natural, human-like follow-up questions.

2. The Memory: Vector Retrieval & Re-ranking (Supabase pgvector)

• Semantic Product Search: Over 300 real fashion products are embedded into a high-dimensional vector space using local sentence-transformers (all-MiniLM-L6-v2).
• Supabase pgvector: Cloud-hosted PostgreSQL handles the vector similarity search (cosine similarity), matching user intent to product meanings rather than exact keyword matches.
• FlashRank Re-ranking: A crucial enterprise addition. We fetch the top 15 results from Supabase, but use an in-process FlashRank cross-encoder to re-rank and extract the absolute top 3 most relevant products before sending them to the LLM.

3. The Interface: Real-time Frontend (Next.js & Vercel)

• Next.js 15 (App Router): Provides robust Server-Side Rendering (SSR) and seamless API route handling.
• Server-Sent Events (SSE): Consumes the FastAPI backend stream to render the LLM's response token-by-token (ChatGPT style).
• Glassmorphic UI: Built with Tailwind CSS and Framer Motion for a premium, dark-mode aesthetic featuring micro-animations and responsive product cards.
• Edge Image Proxying: A custom Next.js API route transparently proxies external CDN images to bypass strict CORS and hotlinking protections.

4. The Engine: Scalable Backend (FastAPI & Render)

• FastAPI / Uvicorn: A high-performance Python web framework handling concurrent async SSE streams.
• Render Web Services: The backend is deployed as a robust web service, automatically building and scaling the Python environment upon every GitHub commit.

---

🔄 System Flow Diagram

graph TD
    A[User Browser - Vercel] -->|POST /chat SSE| B(FastAPI Backend - Render)
    
    B --> C{LangGraph Router Node}
    
    C -->|Classify: SALES| D[Sales Agent]
    C -->|Classify: SUPPORT| E[Support Agent]
    C -->|Classify: OTHER| F[Out of Scope Agent]
    
    D -->|Query Vector| G[(Supabase pgvector)]
    E -->|Query Vector| G
    
    G -.->|Top 15 Matches| H[FlashRank Re-ranker]
    H -.->|Top 3 Matches| I((Groq Llama-3 LLM))
    
    F --> I
    E -.-> I
    
    I -->|Stream Chunks| B
    B -->|Stream JSON| A

Loading

---

🛠️ Local Setup & Development

Prerequisites

• Node.js 18+, Python 3.11+
• A Supabase project with pgvector extension enabled
• A Groq API Key

1. Clone & Set Up Backend

git clone https://github.com/juni2003/NexusAI-Fashion-Assistant.git
cd NexusAI-Fashion-Assistant/backend

# Create and activate virtual environment
python -m venv venv
source venv/bin/activate      # Linux/Mac
.\venv\Scripts\activate       # Windows

# Install dependencies
pip install -r requirements.txt

# Configure Secrets
cp .env.example .env
# Edit .env with your GROQ_API_KEY, SUPABASE_URL, and SUPABASE_KEY

# Launch the API
uvicorn main:app --reload --port 8000

API documentation automatically generated at http://localhost:8000/docs

2. Set Up Frontend

cd ../frontend
npm install

# Connect frontend to local backend
echo "NEXT_PUBLIC_API_URL=http://127.0.0.1:8000" > .env.local

# Launch Next.js
npm run dev

Application live at http://localhost:3000

---

🚀 Deployment Guides

Deploy Backend to Render (Python Environment)

1. In Render, create a new Web Service.
2. Connect this GitHub repository.
3. Configuration:
   • Root Directory: backend
   • Environment: Python 3
   • Build Command: pip install -r requirements.txt
   • Start Command: uvicorn main:app --host 0.0.0.0 --port $PORT
4. Add Environment Variables: GROQ_API_KEY, SUPABASE_URL, SUPABASE_KEY, and PYTHON_VERSION=3.11.0.

Deploy Frontend to Vercel

1. In Vercel, add a New Project and import this repository.
2. Configuration:
   • Root Directory: frontend
   • Framework Preset: Next.js
3. Add Environment Variables:
   • NEXT_PUBLIC_API_URL: Your live Render backend URL (e.g., https://nexusai-backend.onrender.com).
4. Click Deploy.

---

👨‍💻 Author & License

Designed and developed to showcase modern Agentic AI workflows in e-commerce.

This project is open-source and available under the MIT License.