📦 AutoInventory

A Fully Automated, Intelligent Warehouse Management System

---

Project Description

AutoInventory is an end-to-end, event-driven warehouse management solution that addresses real-world FMCG inventory challenges. The system integrates real-time data streaming, predictive analytics, and interactive dashboards to optimize inventory levels, prevent stockouts, and reduce waste—all with zero manual intervention.

---

Dataset

This project uses a simulated FMCG dataset, enriched with:

• Sales Metrics: Sales volume, price, and promotion data.
• Inventory Metrics: Stock levels and replenishment lead times.
• Temporal Components: Date, weekday, and month.
• Geospatial Information: Store locations and product categories.
• Dataset Link: FMCG Sales Demand Forecasting Dataset on Kaggle

---

Architecture and Technologies Used

Architecture

1. Data Streaming: Kafka (running in Docker) streams inventory data into S3.
2. Data Laking: AWS S3 acts as a centralized data lake, storing raw and processed data for further analysis.
3. ETL Pipeline: Glue processes data, and Lambda triggers transformations dynamically.
4. Data Warehousing: Redshift stores analytical-ready data.
5. Dashboards: Streamlit and Power BI provide interactive visualizations.

---

Project Implementation

Step 1: Real-Time Data Streaming

• Technology Used: Apache Kafka (deployed with Docker)
• Simulates real-time inventory updates by streaming data from sales points to AWS S3 as a centralized data lake.
• Key Features:
  • Ensures idempotency with custom logic.
  • Handles hourly batching for efficient ingestion.

Step 2: ETL Pipeline

• Technologies Used: AWS Glue and Lambda
• ETL Highlights:
  • Extracts raw data from S3 and applies transformations like revenue, cost, and profit margin calculations.
  • Processes and loads transformed data into AWS Redshift for analytical queries.
• Event-Driven: Automatically triggered by S3 file uploads using Lambda.

Step 3: Predictive Analytics

• Technology Used: PyTorch custom Neural Network Models
• Forecasts sales and stock levels 5–7 days into the future.
• Key Features:
  • Rolling forecasts for stability.
  • Advanced preprocessing with outlier resistance and missing data handling.
  • Automatic model retraining for evolving data patterns.

Step 4: Dashboards and Insights

• Technologies Used: Streamlit and Power BI

• Streamlit: Provides forecasting, historical data line graphs, and basic insights in real time.

• Power BI: Offers interactive dashboards with drill-down capabilities by store location and product category, and advanced visuals for current stock levels, replenishment times, and revenue trends.
  

---

Technologies

• Data Streaming: Apache Kafka (Dockerized)
• AWS S3: Acts as a centralized data lake for raw and processed data.
• AWS Lambda: For triggering ETL processes dynamically based on S3 events.
• ETL:
  • AWS Glue: For scalable data transformation and data loading to AWS Redshift.
• Data Warehousing: AWS Redshift
• Machine Learning: PyTorch (custom neural network models for time-series forecasting)
• Dashboards: Streamlit and Power BI
• Cloud Management: AWS IAM for permissions and AWS Secrets Manager for secure credential handling
• Experiment Tracking: MLflow
• Database Interaction: SQLAlchemy
• Containerization: Docker

---

Business Impact

• Prevented Stockouts: Predictive analytics keep shelves stocked with high-demand products.
• Reduced Waste: Optimized inventory minimizes spoilage and overstock.
• Improved Decision-Making: Automated KPIs like profit margins and replenishment times enable smarter choices.

---

Connect with Me

Have questions or want to collaborate? Let’s connect!

• LinkedIn: Profile