An end-to-end intelligent analytics platform that processes 100K+ real-world e-commerce transactions, applies ML models for revenue forecasting and customer segmentation, and serves insights through an interactive dashboard deployed on Google Cloud.
Live Dashboard: https://analytics-dashboard-169375435948.asia-south1.run.app
Dashboard Image
InsightX takes raw e-commerce data from the Olist Brazilian Marketplace and transforms it into actionable business intelligence through a fully cloud-native pipeline:
| Stage | What Happens | GCP Service |
|---|---|---|
| Ingest | 9 CSV files (100K+ orders) uploaded to cloud storage | Cloud Storage |
| Warehouse | Raw data loaded and transformed into 7 analytics tables | BigQuery |
| Model | ARIMA+ revenue forecast + K-Means customer segmentation trained in SQL | BigQuery ML |
| Serve | 6-page interactive Streamlit dashboard with dark neon theme | Cloud Run |
| Build | Dockerized app built and deployed via CI pipeline | Cloud Build |
| Develop | All development done on a cloud VM | Compute Engine |
- ₹15.5M total revenue from 96K+ delivered orders across Brazil
- Credit card installment purchases drive 74% of revenue — average 3-4 installments per order
- Single-purchase customers dominate — most customers buy only once (typical marketplace pattern)
- São Paulo generates 40%+ of all revenue — massive geographic concentration
- Revenue forecast shows continued growth with seasonal patterns identified by ARIMA+
Headline KPIs (revenue, orders, customers, sellers, avg order value, review score) with monthly trend charts and year-over-year comparisons.
Monthly revenue time series with AI-generated 6-month ARIMA+ forecast including 90% confidence intervals.
Interactive scatter plot of 4 ML-identified customer clusters (Champions, High Value, Growing Regulars, Casual Buyers) based on spending behavior, with segment-level statistics.
Customer retention heatmap grouped by first-purchase month, showing how many customers return in subsequent months.
Revenue breakdown by payment method (credit card, boleto, voucher, debit) with installment pattern analysis.
Interactive business simulation tool with 4 sliders (price change, marketing spend, discount rate, retention improvement) that projects the impact on revenue, customers, and average order value using economic elasticity models.
┌─────────────┐ ┌──────────────┐ ┌─────────────┐
│ Kaggle │────▶│ Cloud Storage │────▶│ BigQuery │
│ (9 CSVs) │ │ (Data Lake) │ │ (Warehouse) │
└─────────────┘ └──────────────┘ └──────┬──────┘
│
┌───────────┴───────────┐
│ │
┌─────▼─────┐ ┌─────▼─────┐
│ BigQuery │ │ Analytics │
│ ML Models │ │ Tables │
│ (ARIMA+, │ │ (7 tables) │
│ K-Means) │ └─────┬─────┘
└─────┬─────┘ │
│ │
└───────────┬───────────┘
│
┌─────▼─────┐
│ Streamlit │
│ Dashboard │
│ (6 pages) │
└─────┬─────┘
│
┌───────────┴───────────┐
│ │
┌─────▼─────┐ ┌─────▼─────┐
│ Docker │ │ Cloud Run │
│ Container │────────▶ │ (Live URL) │
│(Cloud Build)│ └───────────┘
└───────────┘
| Category | Tools |
|---|---|
| Language | Python 3.10, SQL |
| Dashboard | Streamlit 1.56, Plotly 5.18 |
| Data | Pandas, NumPy, PyArrow |
| ML Models | BigQuery ML (ARIMA+, K-Means) |
| Cloud | GCP — Compute Engine, Cloud Storage, BigQuery, Cloud Build, Cloud Run |
| Deployment | Docker, gcloud CLI |
| Design | Dark neon theme, glassmorphism, Outfit + JetBrains Mono fonts |
InsightX/
├── app.py # Main Streamlit dashboard (6 pages)
├── Dockerfile # Container config for Cloud Run
├── requirements.txt # Python dependencies
├── .streamlit/
│ └── config.toml # Streamlit dark theme config
├── sql/
│ ├── 01_analytics_tables.sql # 7 BigQuery analytics table queries
│ └── 02_bigquery_ml.sql # ARIMA+ and K-Means model training
├── docs/
│ └── architecture.md # Detailed architecture notes
└── README.md
- Google Cloud account with billing enabled
gcloudCLI installed- Python 3.10+
gcloud auth login
gcloud config set project YOUR_PROJECT_ID
gcloud services enable bigquery.googleapis.com storage.googleapis.com run.googleapis.com cloudbuild.googleapis.com- Download the Olist dataset from Kaggle
- Upload CSVs to Cloud Storage:
gsutil cp *.csv gs://YOUR_BUCKET/raw/ - Load into BigQuery via Console (auto-detect schema)
- Run queries in
sql/01_analytics_tables.sqlto create analytics tables - Run queries in
sql/02_bigquery_ml.sqlto train ML models
pip install -r requirements.txt
export GCP_PROJECT=YOUR_PROJECT_ID
streamlit run app.pygcloud builds submit --tag gcr.io/YOUR_PROJECT_ID/analytics-dashboard --timeout=900
gcloud run deploy analytics-dashboard \
--image gcr.io/YOUR_PROJECT_ID/analytics-dashboard \
--platform managed \
--region asia-south1 \
--allow-unauthenticated \
--memory 1Gi \
--set-env-vars GCP_PROJECT=YOUR_PROJECT_IDSource: Olist Brazilian E-Commerce (Kaggle)
| Table | Records | Description |
|---|---|---|
| orders | ~100K | Order metadata and timestamps |
| order_items | ~112K | Line items with price and freight |
| customers | ~99K | Customer demographics and location |
| products | ~32K | Product catalog and dimensions |
| order_reviews | ~100K | Customer review scores and comments |
| order_payments | ~103K | Payment type and installments |
| sellers | ~3K | Seller information |
| geolocation | ~1M | Latitude/longitude data |
| category_translation | 71 | Portuguese to English category names |
Harsh Palyekar — B.Sc. Data Science, Goa University LinkedIn · GitHub