📌 Project Overview

Retail Insights Assistant is a production-ready GenAI chatbot that enables business users to ask natural language questions about Amazon sales data and receive data-driven insights instantly.

👉 Live Streamlit App: https://assignmentgenaiweek4.streamlit.app/

The system combines:

• Python + Pandas for analytics
• Google Gemini Pro API for natural language understanding
• Streamlit for an interactive UI
• A scalable architecture design capable of handling 100GB+ datasets

---

🎯 Objectives

• Build a GenAI chatbot for business analytics

• Answer questions like:

  • Which region performed best?
  • What are top-selling categories?
  • What is the cancellation rate?

• Generate summaries and reports

• Propose a scalable architecture for large datasets

---

🧠 Key Features

✅ Natural-language business queries ✅ Automated sales summaries ✅ Revenue & performance analytics ✅ Interactive Streamlit UI ✅ Gemini AI integration ✅ Scalable design for big data (100GB+)

---

🗂️ Actual Project Structure

week4/
│
├── app.py
├── streamlit_app.py          # Main application
├── Cleaned_Amazon_Sale_Report.csv
├── sales_report.json         # Precomputed analytics summary
├── requirements.txt
├── README.md
├── Assignment.txt
├── Scalability Architecture for 100GB.md
│
├── 1.png   # Chat Interface
├── 2.png   # Example AI Answer
├── 3.png   # Sales Summary Output
├── 4.png   # Data Insights Visualization
├── 5.png   # Architecture / Scalability Diagram

---

⚙️ Tech Stack

Layer	Technology
Language	Python
AI Model	Google Gemini Pro
UI	Streamlit
Analytics	Pandas
Visualization	Streamlit charts
Big Data (Design)	Spark, BigQuery
Deployment (Design)	Kubernetes

---

🏗️ Architecture:

🚀 How to Run the Project

1️⃣ Activate Virtual Environment

PS> & "C:\Users\SaisrisatyaPadala\Desktop\All Stars Training\week4\week4\Scripts\Activate.ps1"

---

2️⃣ Install Dependencies

pip install -r requirements.txt

---

3️⃣ Set Gemini API Key

Create a .env file or set environment variable:

GEMINI_API_KEY=AIzaSy*****

---

4️⃣ Run the Application

streamlit run streamlit_app.py

App opens at: 👉 http://localhost:8501

---

📊 Dataset Used

File: Cleaned_Amazon_Sale_Report.csv

Key Columns:

• Order ID
• Date
• Status
• Category
• Amount
• Quantity
• City, State
• Fulfilled By
• Promotion IDs
• B2B Flag

The dataset is cleaned, normalized, and analysis-ready.

---

💬 Example Questions Supported

• Which state generated the highest revenue?
• What is the total revenue for April 2022?
• Which category sold the most?
• What is the cancellation rate?
• Compare B2B vs B2C sales
• Show monthly revenue trends

---

🖼️ Screenshots Explanation (IMPORTANT)

📸 Screenshot 1 — 1.png

Chat Interface

Shows the main Streamlit UI where:

• User enters business questions
• AI chatbot interaction happens
• Clean and intuitive design

---

📸 Screenshot 2 — 2.png

AI-Generated Answer

Demonstrates:

• Gemini AI answering a business question
• Clear, structured, business-friendly response
• Metrics-based explanation

---

📸 Screenshot 3 — 3.png

Sales Summary Output

Displays:

• Total revenue
• Total orders
• Cancelled orders
• Top states and categories

This summary is generated programmatically and fed to the AI.

---

📸 Screenshot 4 — 4.png

Data Insights & Analytics

Shows:

• Aggregated insights
• Tabular or chart-based analysis
• Useful for decision-makers

---

📸 Screenshot 5 — 5.png

Data Insights

Illustrates:

• Cloud-based data ingestion
• Spark processing
• BigQuery storage
• AI query pipeline
• Caching & retrieval layers

---

🚀 Scalability Challenge: Handling 100GB+ Sales Data

When the dataset grows to 100GB or more, a local CSV + Pandas approach is no longer feasible. The system is redesigned using cloud-native big data and AI retrieval patterns.

---

📥 Data Preprocessing

• Store raw files in cloud object storage (Google Cloud Storage / AWS S3)

• Use Apache Spark for distributed preprocessing:

  • Data cleaning & validation
  • Partitioning by Year, Month, Region
  • Aggregation of daily and monthly metrics

• Save processed data in Parquet / Delta format for efficient reads

---

🗄️ Data Storage & Indexing

• Load processed data into BigQuery / Snowflake

• Use:

  • Partitioned tables (by date)
  • Clustered columns (state, category)

• Create:

  • Aggregated fact tables (revenue, orders, cancellations)
  • Dimensional tables (product, geography)

---

🔍 Efficient Data Retrieval

• Use SQL-based aggregation for structured queries

• Use LangChain + Retrieval-Augmented Generation (RAG):

  • Convert summaries into embeddings
  • Store embeddings in FAISS / Pinecone

• Retrieval Flow:

User Query → Intent Detection
          → SQL / Vector Search
          → Relevant Summary
          → Gemini AI Response

---

☁️ Tools & Technologies Used

Layer	Tools
Storage	GCS / S3
Processing	Apache Spark / Databricks
Analytics	BigQuery / Snowflake
AI	Gemini Pro
Search	FAISS / Pinecone
Orchestration	Airflow
Deployment	Docker + Kubernetes
Caching	Redis

---

✅ Benefits of This Approach

• Scales horizontally to 100GB+
• Fast query response times
• Cost-efficient storage
• AI answers remain accurate and grounded
• Production-ready architecture

---

Superior Architecture

🏗️ Scalability: Handling 100GB+ Data

❌ Problem with CSV + Pandas

• Memory limits
• Slow queries
• Not production-ready

---

✅ Proposed Scalable Architecture

Data Ingestion

• Cloud Storage (GCS / S3)
• Batch ingestion

Processing

• Apache Spark / Databricks
• Daily & monthly aggregations

Storage

• BigQuery / Snowflake
• Partitioned tables

AI Query Flow

User Question
   ↓
Intent Detection
   ↓
SQL / Aggregation Query
   ↓
Summary Data
   ↓
Gemini AI
   ↓
Final Answer

Caching

• Redis for frequent queries

Deployment

• Docker + Kubernetes
• Horizontal Pod Autoscaling

---

🧪 Why This Project Is Production-Ready

✔ Clean architecture ✔ Clear separation of AI & analytics ✔ Prompt-engineering included ✔ Scales from CSV → Big Data ✔ Real business use case ✔ Fully explainable

---

📦 Deliverables Checklist

✅ Working GenAI chatbot ✅ Cleaned dataset ✅ Screenshots (1–5) ✅ Scalability architecture document ✅ Complete README ✅ Ready for evaluation

---

🏁 Conclusion

The Retail Insights Assistant demonstrates how Generative AI + Analytics can transform raw sales data into actionable business intelligence, while being scalable, maintainable, and production-ready.