Real-time shelf analytics, demand forecasting, and computer vision — from edge device to cloud dashboard.
Shelf IQ is built on a fully serverless-first, scalable AWS infrastructure. Every layer — from edge inference to cloud intelligence — is purpose-mapped to the right AWS service.
| AWS Service | Role in Shelf IQ |
|---|---|
| Amazon EC2 | Hosts the production backend API server and processes high-throughput retail event streams |
| Amazon SageMaker | Trains and deploys demand forecasting models (Prophet + LSTM) and shelf anomaly detectors; manages model versioning and A/B evaluation pipelines |
| Amazon Rekognition | Real-time demographic analysis and customer foot-traffic detection from in-store camera feeds; powers the heatmap and engagement scoring engine |
| Amazon S3 | Persistent storage for shelf images, model artifacts, historical sales data, and audit logs with lifecycle tiering |
| Amazon Bedrock | Powers the AI Retail Copilot with foundation model inference (Claude); handles natural language queries, insight generation, and report summarization |
| Bedrock Agents | Orchestrates multi-step agentic workflows — auto-replenishment decisions, cross-sell recommendation pipelines, and what-if scenario planning |
| Amazon SNS | Delivers real-time stock-out alerts, critical inventory warnings, and performance anomaly notifications to store managers and operations teams |
| Amazon DynamoDB | Low-latency NoSQL store for real-time shelf state, SKU metadata, and live sensor readings from edge devices |
| AWS IoT Core | Manages MQTT communication between Arduino edge devices and the cloud backend; handles device registry, telemetry ingestion, and OTA updates |
| Amazon CloudWatch | End-to-end observability — logs, metrics, dashboards, and alarms for both cloud services and edge device health |
| AWS Lambda | Event-driven compute for image processing triggers, alert routing, and lightweight inference tasks at the edge-to-cloud boundary |
| Amazon API Gateway | Unified REST & WebSocket API layer connecting the React frontend, mobile clients, and third-party POS/ERP integrations |
This repository is organized into three branches, each representing a distinct deployment layer:
The React + Vite web application. All 10 dashboard pages, interactive charts, the AR shelf simulator, and the AI Copilot chat interface. This is what store managers and executives interact with daily.
The Python backend hosted on AWS EC2, orchestrating:
- SageMaker model inference endpoints
- Bedrock Agent workflows for autonomous retail decisions
- Rekognition pipelines for demographic and traffic analysis
- REST & WebSocket APIs via API Gateway
- SNS alert dispatching and DynamoDB state management
Firmware and embedded code for in-store edge hardware:
- Arduino-compatible camera modules for shelf image capture
- On-device pre-processing and feature extraction before cloud upload
- MQTT telemetry to AWS IoT Core
- Real-time stock-out detection at sub-second latency
- OTA firmware update support via AWS IoT Device Management
┌─────────────────────────────────────────────────────────────────┐
│ RETAIL STORE │
│ │
│ 📷 Arduino Camera → Edge Inference → AWS IoT Core (MQTT) │
│ 📡 Shelf Sensors → Pre-processing → AWS IoT Core (MQTT) │
└───────────────────────────────┬─────────────────────────────────┘
│
┌───────────▼───────────┐
│ API Gateway │
│ (REST + WebSocket) │
└───────────┬───────────┘
│
┌─────────────────────┼──────────────────────┐
│ │ │
┌───────▼──────┐ ┌──────────▼────────┐ ┌────────▼───────┐
│ EC2 Backend │ │ Lambda Functions │ │ SageMaker │
│ (API + BL) │ │ (Event-Driven) │ │ (ML Models) │
└───────┬──────┘ └──────────┬────────┘ └────────┬───────┘
│ │ │
┌───────▼─────────────────────▼──────────────────────▼───────┐
│ AWS Data & AI Services │
│ S3 (Storage) │ DynamoDB (State) │ Rekognition (Vision) │
│ Bedrock + Bedrock Agents (LLM) │ SNS (Alerts) │
└───────────────────────────────────────────────────────────┘
│
┌───────────▼───────────┐
│ Shelf IQ Dashboard │
│ (React + Vite) │
└───────────────────────┘
Real-time KPI cards (Revenue, Sales Efficiency, Stock-Out Rate), interactive revenue trend charts, multi-store comparison analytics, and an AI recommendations panel with confidence scores — all streaming live from the backend.
Performance tables with color-coded health badges, shelf utilization heatmaps generated from Rekognition camera feeds, and AI placement suggestions ranked by projected revenue impact.
Multi-tier stock alerts (Critical / Warning / Overstock), depletion velocity charts, auto-reorder timelines powered by Bedrock Agents, and SKU risk probability scoring from SageMaker.
Dual-model forecasting engine using Prophet (seasonality) and LSTM (deep learning) trained on SageMaker. Seasonal impact calendar, external factor analysis (weather, local events, festivals), and AI-narrative insight cards.
Association rule mining with Apriori & FP-Growth, combo optimization suggestions, top-performing bundle tracking, and lift/confidence metric breakdowns.
A conversational interface powered by Amazon Bedrock (Claude). Ask natural language questions, receive auto-generated charts and insights, and execute complex multi-step analyses via Bedrock Agents — all with traceable confidence scores.
Interactively simulate price changes, shelf repositioning, and promotional scenarios. Bedrock Agents compute projected revenue, visibility, and demand impact in real time.
Visual drag-and-drop 3D shelf interface. Rearrange SKUs and instantly see predicted visibility scores, profit projections, and traffic flow — all backed by Rekognition spatial analysis.
Multi-store leaderboard, radar chart comparisons, and underperforming store diagnostics with AI-generated turnaround recommendations.
| Layer | Technology |
|---|---|
| Frontend | React 18, Vite, Recharts, Lucide React |
| Backend | Node.js / Python, AWS EC2, API Gateway |
| ML / AI | Amazon SageMaker, Amazon Bedrock, Bedrock Agents |
| Computer Vision | Amazon Rekognition, Arduino OV2640 Camera |
| Edge Hardware | Arduino (ESP32/AVR), AWS IoT Core, MQTT |
| Database | Amazon DynamoDB, Amazon S3 |
| Messaging | Amazon SNS, AWS Lambda |
| Observability | Amazon CloudWatch |
| Role | Primary Value |
|---|---|
| CEO / Executive | High-level KPIs, store rankings, and AI-generated executive summaries |
| Store Manager | Real-time stock alerts, shelf health scores, and replenishment automation |
| Category Manager | Demand forecasting, cross-sell opportunities, and promotional simulation |
| Operations / IT | Edge device management, integration health, and system observability |