Shashank Shandilya

Amazon India – A Decade of Sales Analytics (2015–2025)

Domain: E-Commerce Analytics | Business Intelligence | Data Engineering

Designed and deployed a comprehensive end-to-end e-commerce analytics platform using nearly 1 million Amazon India transactions spanning 10 years (2015–2025). The project transformed raw, messy transactional data into a fully integrated SQL-powered Business Intelligence system with multi-page interactive dashboards.

Key Highlights:

• Engineered advanced data cleaning pipeline resolving 25% real-world data quality issues (date inconsistencies, price formatting errors, city standardization, duplicate handling, outlier correction).
• Performed 20+ Exploratory Data Analysis visualizations covering revenue growth, seasonality, RFM segmentation, price-demand correlation, payment evolution (UPI growth), and festival sales impact.
• Designed optimized SQL database schema (transactions, products, customers, time dimension) with indexing and aggregation queries for dashboard performance.
• Built multi-module Streamlit application including:
  • Executive Business Dashboard
  • Revenue & Growth Analytics
  • Customer Segmentation (RFM, CLV, Retention)
  • Product & Brand Performance Analysis
  • Operations & Delivery Analytics
  • Advanced Predictive & Forecasting Insights
• Implemented pricing analysis, discount effectiveness study, and demand elasticity modeling.
• Integrated forecasting and business health monitoring for strategic decision-making.

Technical Stack:
Python (Pandas, NumPy, Matplotlib, Seaborn, MLflow), SQL (SQLAlchemy), Streamlit, Data Modeling, Statistical Analysis, Business Intelligence

Business Impact Simulated:
• Revenue trend forecasting & growth analysis (2015–2025)
• Prime membership behavior insights
• Customer retention & churn signals
• Category-level profitability & market share evolution
• Payment method evolution analysis (UPI vs COD shift)
• Operational efficiency & delivery performance tracking

🔗 View Live Dashboard

EMIPredict AI – Intelligent Financial Risk Assessment Platform

Domain: FinTech | Machine Learning | Risk Analytics

Designed and deployed a production-ready financial risk analytics platform that evaluates EMI eligibility and predicts maximum affordable EMI using machine learning models trained on 400,000 structured financial profiles. The system integrates end-to-end ML lifecycle management with real-time deployment through an interactive multi-page Streamlit application.

Key Highlights:

• Built a dual-model ML architecture solving:
  • Classification: EMI Eligibility (Eligible / High Risk / Not Eligible)
  • Regression: Maximum Safe Monthly EMI Prediction
• Engineered 22+ financial and demographic features including debt-to-income ratio, credit utilization, employment stability, repayment capacity, and risk scoring metrics.
• Performed model comparison across Logistic Regression, Random Forest, and XGBoost using Accuracy, F1-Score, ROC-AUC (classification) and RMSE, MAE, R² (regression).
• Implemented MLflow for structured experiment tracking, hyperparameter logging, model artifact storage, and version-controlled model registry.
• Designed data preprocessing pipeline including handling missing values, encoding categorical variables, scaling, and validation checks for financial data integrity.
• Built multi-page Streamlit application featuring:
  • Real-Time EMI Eligibility Prediction
  • Affordability Simulation Engine
  • Model Performance Dashboard
  • Risk Category Visualization
  • Deployment-Ready Scoring Interface
• Integrated automated inference pipeline enabling scalable decision support for digital lending environments.

Technical Stack:
Python (Pandas, NumPy, scikit-learn, XGBoost), MLflow (Experiment Tracking & Model Registry), Feature Engineering, Financial Risk Modeling, Data Preprocessing Pipelines, Streamlit Deployment

Business Impact Simulated:
• Automated loan eligibility assessment reducing manual underwriting effort
• Standardized risk-based decision framework for EMI approval
• Real-time borrower affordability analysis
• Improved credit risk transparency through model explainability
• Scalable architecture supporting high-volume digital lending platforms

🔗 View Live Application

PatrolIQ – Smart Safety Analytics Platform

Domain: Public Safety | Urban Analytics | Unsupervised Machine Learning

Designed and deployed a large-scale urban crime intelligence platform analyzing 500,000 sampled records from 7.8 million Chicago crime incidents (2001–2025). The system leverages unsupervised machine learning, geospatial analytics, and dimensionality reduction techniques to identify high-risk zones, temporal crime patterns, and clustering behavior to support data-driven public safety decision-making.

Key Highlights:

• Engineered preprocessing pipeline on 22+ structured crime features including geographic coordinates, timestamps, offense categories, arrest indicators, and severity metrics.
• Implemented multiple clustering techniques:
  • K-Means for structured hotspot segmentation
  • DBSCAN for density-based anomaly zone detection
  • Hierarchical Clustering for nested crime pattern grouping
• Applied PCA and t-SNE for dimensionality reduction, compressing high-dimensional crime features into 2–3 principal components explaining 70%+ variance.
• Evaluated clustering performance using Silhouette Score and Davies–Bouldin Index to ensure cluster separation quality.
• Built interactive multi-page Streamlit dashboard including:
  • Geographic Heatmaps & Hotspot Visualization
  • Hourly & Seasonal Crime Trend Analysis
  • Cluster Comparison Dashboard
  • Dimensionality Reduction Visual Explorer
• Integrated MLflow for experiment tracking and clustering model comparison.

Technical Stack:
Python (Pandas, NumPy, scikit-learn), K-Means, DBSCAN, Hierarchical Clustering, PCA, t-SNE, MLflow, Geospatial Data Analysis, Streamlit Deployment

Business Impact Simulated:
• Identified high-risk crime zones for optimized patrol allocation
• Discovered peak crime time windows (late-night high-frequency patterns)
• Enabled evidence-based public safety planning
• Reduced manual crime pattern analysis through automated clustering
• Demonstrated scalable analytics architecture for smart city applications

🔗 View Live Application

SmartVision AI – Multi-Class Object Recognition System

Domain: Computer Vision | Deep Learning | Real-Time AI Systems

Designed and deployed a production-ready computer vision platform integrating transfer learning-based image classification and YOLOv8 multi-object detection. The system was trained on a curated 25-class subset of the COCO dataset (2,500+ images) and optimized for real-time GPU-accelerated inference using CUDA.

Key Highlights:

• Implemented and fine-tuned multiple CNN architectures using transfer learning:
  • VGG16 – Baseline model evaluation
  • ResNet50 – Improved feature extraction depth
  • MobileNetV2 – Lightweight optimized inference
  • EfficientNet-B0 – Achieved highest classification accuracy (88–93%)
• Fine-tuned YOLOv8 for real-time multi-object detection achieving:
  • mAP@0.5: 85–90%
  • 30–50 FPS GPU inference performance
• Engineered complete inference pipeline:
  • Image Upload → YOLO Detection → Optional CNN Verification
  • Non-Maximum Suppression (NMS)
  • Confidence Threshold Filtering
  • Bounding Box Rendering with Class Labels & Scores
• Built multi-page Streamlit application including:
  • Image Classification Dashboard
  • Object Detection Interface
  • Live Webcam Detection (Real-Time Inference)
  • Model Performance Analytics
  • Technical Architecture Documentation
• Optimized model training and inference using CUDA-enabled NVIDIA RTX 2060 GPU, reducing inference latency significantly.
• Evaluated models using Accuracy, Precision, Recall, F1-Score, Confusion Matrix, mAP metrics, inference time benchmarking, and model size comparison.

Technical Stack:
Python, PyTorch, Torchvision, Ultralytics YOLOv8, Transfer Learning, CNN Architectures (VGG16, ResNet50, MobileNetV2, EfficientNet), OpenCV, Data Augmentation, GPU Training (CUDA), Streamlit Deployment

Business Applications:
• Smart City surveillance & traffic analytics
• Retail shelf monitoring & product recognition
• Security & automated compliance monitoring
• Healthcare PPE detection & safety validation
• Logistics & warehouse automation systems

🔗 View Live Application

AgriVision-Bridge 2026 – Multi-Modal Crop Diagnostic AI System

Domain: AgriTech | Computer Vision | Generative AI

Designed and deployed a multi-modal AI platform that combines YOLO-based crop disease detection with a locally deployed LLM (Llama 3 via Ollama) to generate explainable, structured diagnostic reports. The system processes crop leaf images in real time and delivers actionable treatment recommendations for precision agriculture use cases.

System Architecture:

• Vision Layer: YOLO model trained on 10,000+ labeled crop images for multi-class disease detection
• Explainability Layer: Lesion segmentation and visualization using OpenCV (watershed & contour mapping)
• Reasoning Layer: Structured prompt engineering integrated with Llama 3 (local inference via Ollama)
• Deployment Layer: GPU-accelerated Streamlit application with real-time inference

Key Highlights:

• Fine-tuned YOLO model for multi-class plant disease classification with optimized bounding-box detection and confidence scoring.
• Integrated OpenCV-based lesion visualization to highlight infected regions and improve interpretability of AI predictions.
• Designed structured prompt templates to generate context-aware diagnostic summaries including severity assessment and treatment guidance.
• Benchmarked GPU (CUDA) vs CPU inference latency to ensure real-time performance optimization.
• Implemented fully local multi-modal pipeline (Image → Detection → Prompt Structuring → LLM Diagnosis → Report Output).
• Built production-ready Streamlit interface featuring:
  • Image Upload & Real-Time Detection
  • Disease Confidence Scoring
  • LLM-Generated Diagnostic Reports
  • Visualization of Infected Regions
  • Performance Metrics Display

Technical Stack:
Python, Ultralytics YOLO, OpenCV, PyTorch, CUDA (GPU Acceleration), Llama 3 (via Ollama – Local LLM), Prompt Engineering, Streamlit Deployment

Business Impact Simulated:
• Early crop disease detection to prevent yield loss
• AI-driven advisory for farmers and agri-consultants
• Reduced dependency on manual agronomic inspection
• Deployable local AI system suitable for rural environments
• Scalable precision agriculture analytics framework

🔗 View Live Application

Professional Certification – Advanced Programming & Master Data Science

Indian Institute of Technology (IIT) Kanpur | 2025-2026

Industry-aligned certification covering end-to-end Data Science lifecycle including Python programming, SQL (basic to advanced), Probability & Statistics, Exploratory Data Analysis, Machine Learning (Regression, Classification, Clustering, Ensemble Models), Deep Learning with PyTorch, NLP, Computer Vision (YOLO), and Model Deployment on Streamlit, and AWS Cloud.

Hands-on implementation across real-world case studies including fraud detection, customer segmentation, recommendation systems, and deployment-ready AI applications.

Master of Business Administration (MBA) – IT & Analytics

Guru Gobind Singh Indraprastha University (GGSIPU) | 2022 – 2024

Specialized in Information Technology and Analytics with focus on business intelligence, predictive modeling, financial analytics, and data-driven strategy formulation. Developed strong foundation in statistical analysis, forecasting models, and technology-enabled decision systems aligned with enterprise analytics use cases.

Bachelor of Commerce (Honours)

Maharaja Agrasen Institute of Management Studies (GGSIPU) | 2016 – 2019

Specialized in Database Management Systems and Financial Management, building core expertise in accounting systems, structured data handling, financial reporting, and quantitative analysis — forming the analytical base for later data science and risk modeling projects.