Data Science Portfolio

About Me

Data Scientist specializing in supply chain optimization and demand forecasting for enterprise FMCG and logistics clients. Proven expertise in building production-grade ML systems that solve complex business challenges in promotional planning, inventory optimization, and workforce analytics.

Currently developing advanced forecasting models for major clients including Pepsi and Asahi Beverages, with focus on promotional effectiveness analysis and product cannibalization modeling. Experience spans the full ML lifecycle—from exploratory data analysis and feature engineering through to AWS cloud deployment and MLOps pipeline automation.

Technical strengths include time-series forecasting, causal inference methodologies, statistical modeling, and scalable ML pipeline development. Demonstrated ability to translate complex analytical insights into actionable business strategies for retail optimization, supply chain planning, and organizational decision-making.

Master's degree in AI and Machine Learning from the University of Adelaide, with hands-on expertise in modern ML frameworks, cloud infrastructure, and production system observability.

Current Focus

Data Scientist

Complexica

AI/ML and SaaS Solutions

March 2025 - Present

Core Responsibilities & Technical Focus:

Promotional Forecasting & Demand Prediction: Developed advanced time-series forecasting models for major FMCG clients (Asahi Beverages, Carlton and United Breweries) to optimize promotional strategies and inventory planning. Implemented feature engineering pipelines incorporating promotional mechanics, seasonality, and external market factors to improve forecast accuracy.
Cannibalization & Market Basket Analysis: Built statistical models to quantify product cannibalization effects for Asahi's beverage portfolio, enabling data-driven SKU rationalization and promotional planning decisions. Applied causal inference methodologies to isolate true incremental lift from promotional activities.
Employee Churn Prediction (Emite): Designed and deployed machine learning models for workforce analytics, predicting employee attrition risk and identifying retention intervention opportunities for Emite (A Prophecy Solution). Developed interpretable models to surface key churn drivers for HR strategic planning.
Supply Chain Analytics & Optimization: Partnered with enterprise clients to build predictive models for demand forecasting, inventory optimization, and logistics planning. Developed solutions addressing complex multi-echelon supply chain challenges.
Production MLOps & Cloud Deployment: Architected end-to-end ML pipelines on AWS infrastructure with emphasis on scalability, monitoring, and automated retraining. Implemented CI/CD workflows for model deployment, A/B testing frameworks, and comprehensive logging/observability systems for production model performance tracking.
Log Analysis & System Intelligence: Built automated log parsing and anomaly detection systems to identify system issues, optimize performance, and extract actionable insights from application telemetry data for Snare (A prophecy Solution).

Key Technologies & Tools:

Python and Java Programming AWS Time Series Forecasting AI Agents MLOps Causal Inference Supply Chain Optimization Feature Engineering Anomaly Detection Market Basket Analysis

Key Achievements

15%

Revenue Growth

Implemented customer segmentation and acquisition models driving optimized targeting and engagement strategies

70%

Efficiency Improvement

Developed automated data warehousing and analytics pipelines while maintaining high data quality

92%

Query Accuracy

Built and deployed enterprise-scale ML systems processing millions of daily queries

Education

Master of Artificial Intelligence and Machine Learning

University of Adelaide

Sep 2022 - Aug 2024

GPA: 6.2/7.0 (Distinction)

Bachelor of Science in Computer Science and Engineering

North South University

Jan 2015 - Dec 2019

First Class

Publications

Advancing Community Directories: Leveraging LLMs for Automated Extraction in MARC Standard Venue Availability Notes

Australasian Language Technology Association Workshop (ALTA 2024), ACL Chapter

Supervisors and Co-Authors: Dr. Thushari Atapattu, Dr. Menasha Thilakaratne

View Publication →

A Deep Gaussian process for forecasting crop yield and Time-Series analysis of precipitation based in Munshiganj, Bangladesh

IEEE International Geoscience and Remote Sensing Symposium

Supervisor and Co-Author: Dr. Shahnewaz Siddique

View Publication →

EZ Parking: Smart Parking Space Reservation Using Internet of Things

International Conference on Advanced Computer Science and Information Systems

Supervisor and Co-Author: Tamanna Motahar

View Publication →

Case Studies

ExpertEaseAI Enterprise AI Agent

Feb 2024 - July 2024

AI & ML Solutions

Challenge

Develop a robust and scalable chatbot for a diverse range of industries, including healthcare, finance, and e-commerce.

Solution

- Architected a state-of-the-art Multimodal RAG Chatbot integrating multiple LLM APIs including OpenAI, Anthropic, and Llama
– Designed and built sophisticated data extraction pipelines to gather information from diverse sources, including websites, APIs, and documents
– Developed advanced preprocessing algorithms and cleaning methodologies to ensure high-quality, accurate data for model training and inference
– Optimized and fine-tuned language models for specific domains, significantly enhancing response accuracy and relevance in conversational AI applications
- Implemented automated A/B testing workflows reducing experiment cycle time by 60%

60%

Faster Experiments

92%

Model Accuracy

100%

Automation

Technologies Used

LangChain Vector Databases (Chroma) AWS ETL Pipelines Hugging Face Prompt Engineering LLM API Integration RAG Chatbots

Data-Driven Insights for Norwood Council

2024

Data Analysis & Visualization

Norwood, Payneham & St. Peters Council

Challenge

Evaluate user engagement trends and behavioral patterns across various categories and organizations from 2019 to 2023 using Google Analytics and SAcommunity database data.

Solution

- Developed comprehensive data analysis pipelines using Python and Numpy
- Created interactive PowerBI dashboards for stakeholder reporting
- Analyzed trends across 8 key dimensions including user demographics, device usage, and engagement patterns
- Implemented data cleaning and transformation workflows for accurate analysis

Key Achievements

62%

Mobile Usage Growth

38%

Organic Search Increase

125%

Youth Engagement Growth

Key Insights

• Identified significant growth in youth engagement (18-24 age group)
• Revealed strong shift towards mobile device usage
• Discovered patterns in category preferences and user behavior
• Analyzed gender-based engagement trends

View Interactive Dashboard

Read the full Report

Check out my SAcommunity Portfolio

Technologies Used

Python Numpy PowerBI Google Analytics Data Analysis ETL Pipelines

AI-Powered Information Extraction for Community Services

View Publication

2024

Australasian Language Technology Association Workshop

University of Adelaide & SAcommunity

Research Overview

Led the development of an automated system to extract and structure venue availability information from unstructured text into MARC standard format using RoBERTa transformer models. This research directly impacts 10,000+ community organizations across South Australia.

Technical Approach

• Fine-tuned RoBERTa model on community service descriptions
• Implemented comprehensive data preprocessing pipeline
• Developed active learning strategies for efficient data annotation
• Created automated MARC standardization system

78%

Model Accuracy

70%

Manual Work Reduction

92%

User Queries Handled

Key Innovations

• Novel application of LLMs for community information management
• Integration of MARC standards for structured data extraction
• Development of efficient active learning annotation workflow
• Implementation of automated information standardization

Research Impact

This research significantly improved community service information management by automating the extraction and standardization of venue information. The system reduced manual processing time by 70% while maintaining high accuracy, benefiting over 10,000 organizations across South Australia. Published at ALTA 2024 (ACL Workshop), demonstrating successful industry application of state-of-the-art NLP systems.

Technologies Used

RoBERTa Python PyTorch MARC Standards Hugging Face Active Learning

Enterprise Data Transformation for a Major Retail Chain

May 2021 - April 2022

KH Analytics

Dhaka, Bangladesh

Project Overview

Led a comprehensive digital transformation initiative for a major retail chain, implementing data-driven solutions across their omnichannel operations. The project encompassed data warehousing, analytics pipelines, and smart data products, resulting in significant improvements in operational efficiency and revenue growth.

70%

Data Quality Improvement

30%

Operational Efficiency

15%

Revenue Increase

Key Initiatives

Centralized Data Architecture

Designed and implemented a Snowflake-based data warehouse integrating multiple channels (e-commerce, telesales, brick-and-mortar), creating a unified data ecosystem for cross-channel analytics.

Advanced ETL Solutions

Developed robust data pipelines using Python, PySpark, and SQL, improving data quality and accessibility across the organization. Implemented automated quality checks and monitoring systems.

AI-Powered Retail Solutions

Created and deployed multiple smart data products including:
• Personalized recommendation engine
• Dynamic pricing system
• Customer segmentation models

Technologies Used

Snowflake Python PySpark SQL Tableau PowerBI FiveTran ETL PyTorch

Project Impact

• Established enterprise-wide data infrastructure serving multiple business units
• Achieved 70% improvement in data quality and accessibility
• Drove 30% increase in operational efficiency through data-driven decision making
• Generated 15% revenue growth through smart data products
• Created scalable data strategy roadmap for future growth

Economic Well-being Analysis Using Deep Learning & Satellite Data

2021

North South University & Brac University

Bangladesh & India

Research Overview

Developed an innovative approach to estimate economic well-being using deep transfer learning and remote sensing data. The research covered regions across Bangladesh and six Indian states, achieving significant accuracy in predicting economic indicators through satellite imagery analysis.

80%+

Regression Accuracy

39K

Satellite Images

7

Regions Analyzed

Technical Approach

• Implemented VGG19 deep learning architecture for feature extraction
• Combined day and night satellite imagery for comprehensive analysis
• Developed custom data preprocessing pipeline for geospatial data
• Created automated feature extraction workflows

Research Components

Data Integration

Combined multiple data sources including Google Maps static API, night-time satellite imagery, and demographic data to create a comprehensive analysis framework.

Model Innovation

Enhanced VGG19 architecture with custom layers for economic indicator prediction, achieving higher accuracy than existing state-of-the-art approaches.

Statistical Analysis

Conducted comprehensive statistical analysis across regions, revealing significant patterns in economic distributions and demographic variations.

Technologies Used

VGG19 Python Deep Learning Remote Sensing Geospatial Analysis Transfer Learning

Research Impact

This research provided a novel approach to economic analysis using AI and satellite data, offering a cost-effective alternative to traditional economic surveys. The methodology demonstrated high accuracy in predicting economic indicators across diverse geographical regions, contributing to both academic research and potential policy applications.

Featured Projects

View Code

Building AI Agents with LangGraph

Comprehensive tutorial and implementation guide for building sophisticated AI agents using LangGraph. Features a complete fitness assistant example with state management, error handling, and AWS Bedrock integration.

Key Features

• State Management
• Error Handling
• AWS Integration
• Testing Framework

Technologies

LangGraph AWS Python

Complete implementation with best practices

Production-ready code structure

Comprehensive documentation

AI Agents LLMs Tutorial Best Practices AI Development AWS Python LangGraph

View Dashboard

Community Engagement Analytics Dashboard

Interactive PowerBI dashboard developed for Tea Tree Gully Council (FY 2022-2023), providing deep insights into community engagement patterns and visitor demographics through advanced geospatial visualization.

Key Features

• Geotagged organization mapping
• Global visitor tracking
• Session analytics
• Interactive filters

Technologies

PowerBI DAX Google Analytics Geospatial

Organization location mapping with interactive address lookup

Global visitor tracking with geographic distribution visualization

Comprehensive session analytics with temporal trends

Dynamic filtering capabilities for detailed analysis

Data Visualization Analytics Community Engagement Geospatial Analysis

Early Advantage Network - Supply Chain Finance Platform

Led the development of Bangladesh's first Dynamic Discounting Platform, connecting SME suppliers with corporate vendors. Built a comprehensive supply chain finance solution including e-invoicing, inventory management, and real-time analytics for optimizing working capital and supply chain health.

Key Features

• Dynamic Discounting Engine
• E-invoicing Platform
• Inventory Management
• Real-time Analytics Dashboard
• Cash Flow Optimization

Technologies

Django Node.js REST API Chart.js

First Dynamic Discounting Platform in Bangladesh

Comprehensive Business Analytics Dashboard

Real-time Notification System

API-based Microservices Architecture

Supply Chain Finance FinTech E-invoicing Real-time Analytics Inventory Management CI/CD

Key Achievements

• Pioneered Bangladesh's first Dynamic Discounting solution
• Built comprehensive inventory management system
• Developed real-time business analytics dashboard
• Implemented secure API-based backend architecture
• Integrated automated notification and email systems

Customer Acquisition Optimization

Developed ML models for customer segmentation and acquisition strategy optimization. Implemented A/B testing framework for marketing campaigns evaluation, created predictive models for customer lifetime value analysis.

Key Features

• Customer Segmentation
• A/B Testing Framework
• Predictive Modeling
• Automated Reporting

Technologies

Python ML Analytics

15% increase in customer acquisition

70% improvement in campaign efficiency

Machine Learning A/B Testing Analytics

Enterprise AI System for Community Services

Led development of an AI-powered user engagement system serving 10,000+ organizations. Implemented production-grade NLP pipeline reducing response time by 70%.

Key Features

• NLP Pipeline
• User Analytics
• Automated Workflows
• Service Integration

Technologies

LLMs NLP Python

70% reduction in response time

10,000+ organizations served

NLP LLMs Production ML

Real-time Decision Support System

Developed an enterprise-scale customer query system handling 10,000+ daily queries with 92% accuracy. Implemented optimized algorithms for enhanced customer engagement and designed scalable architecture for multi-source data integration.

Key Features

• Real-time Processing
• Query Optimization
• Multi-source Integration
• Automated Response

Technologies

Python SQL Apache Kafka

92% query accuracy rate

10,000+ daily queries processed

Real-time processing capabilities

Real-time Systems Data Integration Query Optimization Scalable Architecture

Satellite Data Analysis

Led development of predictive models using satellite imagery and economic indicators, achieving 89% accuracy in forecasting economic growth patterns. Engineered feature extraction pipelines using CNNs to process large-scale geospatial data.

Key Features

• CNN Architecture
• Feature Extraction
• Geospatial Analysis
• Economic Modeling

Technologies

TensorFlow CNN Python

89% prediction accuracy

Processed 2TB+ satellite data

Multi-region analysis coverage

Computer Vision Deep Learning Geospatial Analysis Economic Forecasting

Skills

Machine Learning & AI

Predictive Analytics

Customer Behavior Modeling

A/B Testing

Growth Metrics

LLMs

Enterprise Solutions

Marketing Automation

Growth Analytics

Business Process Automation

Cloud Architecture

Data Engineering

Python

SQL

Spark

Snowflake

AWS

Analytics & Insights

PowerBI

Tableau

Marketing Dashboards

Growth KPI Monitoring

ML Technologies

TensorFlow

PyTorch

MLflow

A/B Testing Tools

Blog & Articles

January 22, 2025

Building an AI Fitness Agent with LangGraph: A Complete Guide

A step-by-step guide to building sophisticated AI agents using LangGraph. Learn about state management, error handling, AWS integration, and best practices for creating production-ready AI applications. Includes a complete implementation of an AI fitness assistant.

LangGraph AI Development AWS Bedrock Tutorial Best Practices

January 21, 2025

Mastering Prompt Engineering: A Comprehensive Guide for the AI Era

A deep dive into modern prompt engineering techniques, from fundamental concepts to advanced frameworks. Learn about the CO-STAR, CRISPE frameworks, security considerations, and practical implementation strategies for effective AI communication.

AI Engineering LLMs Best Practices Technical Guide Prompt Engineering

April 15, 2024

How to: Download a PBIX file from Power BI Service when the option is greyed out

In this article, I will show you how to transfer PBIX file data between Power BI Service and PowerBI Desktop when the option is greyed out using the PowerBI APIs.

Power BI Data Engineering REST APIs Data Visualization

Dec 14, 2023