Transforming Data
into Strategic
Insights.

Using machine learning to predict which customers are likely to leave the bank. The aim is to recognize which users are probable to churn. We can input a customer data and get the result on whether this customer might stick with the bank or leave for another competitor based on the machine learning model.

The data was trained and tested with 4 ML models. Class imbalance was addressed with SMOTE and compared how weights and resampling affect model perfomance. After careful consideration, Random Forest was used for its robust performance. The model was serialized and deployed to an API endpoint with Fast API and this was connected to a frontend to add interactivity.

The goal of this project was to identify fraudulent credit card transactions using a popular dataset in machine learning. With 30 features of over 200,000 transactions, the dataset was heavily skewed. Over 20,000 rows were removed in the outlier analysis with a distance of 3 * IQR above and below.

Another objective was to try building a model with Auto ML libraries. After doing some research, I have chosen pycaret. The binary classifier was to build several models and find the optimal one; Extra Trees classifier had the best overall performance. Then I used Streamlit to deploy the model and build an interactive interface.

Powered by AI, this bot delivers near real-time market updates, in-depth summaries, and strategic financial insights - empowering you to make informed decisions in the fast-paced world of investing in the Colombo Stock Exchange. 🚀

Features: Real-Time Data & AI Insights, Daily CSE Market Statistics, Strategic Financial Summaries, AI-Powered Q&A, and Multi-Language Support.

End to End Big Data Solution using the Lambda Architecture. This solution is capable of processing both historical and batch data sources by incorporating historical datasets, Near-real-time data with APIs from Crickbuzz and Cricketdata.org and streaming data pushed to event hub that's updated every 15 seconds to simulate ball by ball data.

Technologies used: Azure Data Factory, Azure Databricks, ADLS Gen2, Azure Synapse Analytics, Entra ID, Event hubs, Azure Stream Analytics, Python, Power BI, Azure Cost Management.

Agentic Retrieval-Augmented Generation (RAG) workflow built using n8n, designed to provide information from machine learning-related documents stored in Google Drive. Includes 2 pipelines to generate and create document indexes and then for llm to provide relevant results using that context.

The workflow leverages Pinecone as a vector database, OpenAI Text Embeddings, LLaMA3 70B (via Groq), and Gemini 1.5 Flash for validating and reasoning over vector store queries.

This project leverages machine learning to predict hotel booking outcomes (check-in, cancel, or no-show) for a chain operating airport hotels, resorts, and city hotels. By analyzing historical booking data, the project aims to optimize resource allocation and mitigate financial losses.

The hotel industry faces significant revenue and operational challenges due to booking cancellations and no-shows. This project addresses these issues for Hotel A by: Conducting exploratory data analysis (EDA) to uncover patterns in reservation statuses. Developing predictive models to classify booking outcomes (Check-In, Cancel, No-Show). Deploying a user-friendly application to predict reservation statuses based on booking details. The primary evaluation metric is the F1-score, chosen to handle the dataset's class imbalance, where "Check-In" is the majority class, and "Cancel" and "No-Show" are minority classes. T Random Forest with class balancing and feature selection achieved the highest F1-score of 0.66.

Olist is a major Brazilian Ecommerce marketplace. I used 100,000+ records of real commercial data across 9 relational tables to create a production scale data pipeline.

Technologies: AWS S3, AWS Glue, Amazon Athena, Snowflake, Power BI, AWS IAM, AWS Cloudwatch.

This project delved into the Amazon Prime Movies and TV shows titles and using this dataset to build an end to end production level pipeline.The pipeline follows a medallion architecture (bronze, silver, gold layers): Raw CSV files stored in ADLS (Bronze), Cleaned data in Parquet (Silver), and Aggregated business-ready tables in Delta format (Gold).

The data was first fetched from the github with the copy function in Azure data factory. The Amazon Prime Movies and TV Shows dataset (sourced from Kaggle) contains metadata on titles, including release year, genre, duration, and ratings. the data fetched from github is stored into the bronze container as raw csv files.Then the databricks was connected to ADLS storage using client secrets to access the data in our raw data files in the bronze container . silver layer transformations were done on this data fetched and after doing some EDA and schema transformations, the data was written back to the silver container in the parquet format. Then the data is fetched back again from the silver layer and final preprocessing, creation of new columns such as genre, changing the date format to a consistent format etc were applied. The results were stored back into gold containers in the delta format. The results can then be queried in synapse by using the lake from the gold layer data. Synapse setup is set to use built-in serverless pools

Technologies: Azure Synapse, Azure Databricks, ADLS Gen 2, ADF.

The data source for this project is a real world dataset of all the transactions occurring between 01/12/2010 and 09/12/2011 for a UK-based and registered non-store online retail.The company mainly sells unique all-occasion gifts. Many customers of the company are wholesalers. Features were created for customers that were then used to do the customer segmentation. Then I leveraged BigQuery MKL to create ML models to segment the customers through clustering. After doing somme preliminary data cleaning the BigQuery gold tables along with other tables were connected to Tableau and thus used to create dashboards. Dashboards include both analytics insights as well as ML driven findings.

The results were visualized in Tableau dashboards including both analytics insights and ML driven findings as in the image. click on the github to see more

project is a sales dashboard built using Looker Studio that provides insights into the sales performance of a retail company. This interactive live dashboard is designed to help stakeholders understand key sales metrics, identify trends, and make informed business decisions.

Austin Bike Sharing Dataset is a real world dataset of Bike Sharing Activities in the City of Austin, Texas. The Data was Sourced from BigQuery public datasets. BigQuery thus acts as the data repository and SQL engine for the queries. Visualization tool of choice for this project was chosen to be Grafana.

An image description and visual question answering AI using hugging face models. Bootstrapping Language-Image Pre-training ( BLIP ) models from Salesforce was chosen as the models and the application was build using Gradio. We can easily input an image in our own machine or even use the clipboard and the webcam to get the image. This input is then used by the image captioning model to generate a description for the image as to what it's about. The question answering BLIP model then can interactively answer questions regarding the image which gives short answers.

A simple code converter built using Streamlit and the OpenAI API. The user can input R code and the application will return the equivalent Python code.

The dataset consisted of over 300 records of rainfall and other climate regressor data. The goal was to predict future rainfall probabilities based on patterns identified from this dataset.

The primary objective of this interactive dashboard is to analyze sales trends for Adidas products in the United States.

By leveraging Power BI, this dashboard aims to provide insights into various aspects of Adidas sales, such as product performance, sales channels, geographical distribution, and seasonal variations. Understanding these trends can help stakeholders make informed decisions regarding marketing strategies, inventory management, and expansion opportunities. What are the sales channels driving more sales. more units. Product profit margins and average cost of production for a unit and how these vary across time and location are explored using this dashboard.

he main goal of this interactive dashboard is to highlights trends in different roles and titles. I'm focusing on things like experience, employee residence, and salary information to provide a clear picture of the job market landscape. The underlying reason for selecting this domain was that the job market is dynamic with changes happening in the industry with the evolving market. Job salaries are important for job seekers, employers, and policymakers who want to understand what's happening in the job market. this dashboard analysis of the data science jobs brings several insights about the landscape of jobs in data science.

Human Resources Dashboard for a company called 'Awesome Chocolates'. Provides comprehensive overview of gender distribution, age, job types, headcount growth, and salary analytics.

This project is an AI-powered chatbot designed to help employees quickly find answers to questions about internal company policies and procedures. Instead of going through documents,employees can basically ask the chatbot regarding any questions related to work policies. It uses a Retrieval-Augmented Generation (RAG) architecture to provide accurate and context-aware responses based on a collection of company documents.

A Fully Responsive single page Website design and development featuring a concept Sri Lankan Energy Drink- Enerja.

Web dashboard on Climate Change's Impact on Sri Lanka and the world. Shows the trends of global warming, CO2 emissions over the decades and future trends. This analysis gives unique insights into the Industries that contribute most towards global warming, the countries with the highest rates of emission, the cumulative emission contribution towards global warming by top emitters and attempts to highlight the gravity of the current situation.

Real time updated weather for key Sri Lankan cities show the current weather data for the visitor's time. Sri Lanka's greenhouse emission trends, fuel consumption, mean temperatures and impact of climate change for Sri Lanka are few of the studies done through this project.

A personal finance tracker that allows users to track their spending and income. Features OCR to scan receipts, dark mode, and data export.