With a steadfast dedication to my, I have acquired extensive expertise in the fields of deep learning, machine learning, data engineering and data analysis. Throughout my career, I have consistently honed my skills in these domains, enabling me to deliver robust and innovative solutions to complex problems. Along way, I have had privilege of contributing to a multitude of projects, each of which has played a significant role in shaping my experience. I cordially invite you to explore some of the notable ones below.
Spotify Data Pipeline using Snowflake, AWS, Python
Designed an automated ETL pipeline that extracted data from Spotify API using Python, transformed it with AWS Lambda and Glue and loaded it into Snowflake for efficient data warehousing. Implemented AWS including S3, Glue, lambda for scalable and efficient data pipeline for real-time analytics with Power BI.
​
Tools Used: Python, Snowflake, AWS(S3, Cloud Watch, Lambda)
Streamlit-Based Intelligent Chatbot for Data Interaction
Developed a chatbot using streamlit and Langchain. Implemented prompt engineering and created agents for SQL and Python code generation from natural language. Set up a MYSQL database and integrated LLMs for query generation. Focused on memory, guardrails, and transformer architecture for optimizing LLM performance.
​
Tools Used: Streamlit, Langchain, MySqlDB(Loacal Instance), LLMs(GPT-4)
Dimensional Modelling — Creating Fact & Dimensional Tables (AWS S3, Snowflake dB)
In the world of data warehousing, Dimension Modelling is like designing the ultimate playbook for your favorite sports team. Just as a coach structures the team with offensive and defensive strategies to ensure victory, Dimension Modelling structures data into fact and dimension tables, paving the way for efficient querying and insightful analytics. This blog dives into how to create these all-star tables using AWS S3 and Snowflake DB, two powerhouse players in the cloud data ecosystem, transforming your data game into a championship-winning performance.
Tools Used: SQL, AWS (S3, IAM), Snowflake DB
Information Extraction via text-to-text method
In this project, I used a T5 model for extracting adverse drug events and medication information from text, addressing limitations in existing methods. Using the T5 architecture and tailored data preprocessing, the model achieved robust entity extraction. The model was successfully trained and predicted results on the test dataset, demonstrating significant progress in automating medical text processing.
Tools Used: Python (SimpleT5, pandas, scikit-learn, PyTorch Lightning)
Brain Tumor Segmentation Using U-Net
In this project I used a U-Net model for medical image segmentation, overcoming limitations in existing methods. Using U-Net architecture and modified loss functions, the model achieves impressive metrics: a loss of 0.1164, accuracy of 95.15%, Dice coefficient of 0.3407, and IoU of 0.2059. These results represent substantial advancements, with potential to improve diagnostic accuracy and patient outcomes in medical applications.
​
Tools Used: Python (Tensorflow, Keras, PyTorch, OpenCV, Matplotlib, pandas)
Enhancing face recognition with occlusion handling
This research focuses on improving face recognition accuracy in cases with occlusions like glasses or masks. It evaluates advanced deep learning models, particularly CNNs, Siamese networks, and models like VGGFace and FaceNet. Testing on various datasets, including Labeled Faces in the Wild, shows CNNs achieving the highest accuracy at 95.4%, offering insights for occlusion-aware face recognition.
​
Tools Used: Python (Tensorflow, Keras, PyTorch, OpenCV, Matplotlib, pandas)
Recognition of abnormality in breast thermograms
Project aimed to detect early stage breast cancer using thermograms. Achieved over 92.15% accuracy using deep learning and GradCam visualization. Shows potential for early detection of breast cancer.
​
Tools Used: Python (Tensorflow, Keras, PyTorch, OpenCV, Matplotlib, pandas), AWS
I have worked on an exciting project focused on recognizing face masks using convolutional neural networks (CNNs). This project aimed to address the pressing need for mask detection in various environments and achieved an impressive accuracy rate of 98%.
​
Tools Used: Python (Tensorflow, Keras, PyTorch, OpenCV, Matplotlib, pandas)
Hotel Booking Demand Analysis
This project analyzes hotel booking data to identify factors influencing bookings, cancellations, and guest behaviors. By employing data preprocessing, exploratory data analysis, and visualizations, it uncovers insights on booking patterns, seasonal trends, and pricing strategies. The project introduces a GUI developed with PyQt5, enabling users to preprocess data, conduct analyses, and generate visualizations. Key results include detecting higher bookings in summer, dominant customer types, and trends in cancellations, all aiding in enhanced operational efficiency and customer satisfaction.
Tools Used: Python (Pandas, NumPy, Matplotlib, Seaborn, PyQt5
