Oh, Hello there!
I'm Mayur, a Data Scientist
specialized in creating intelligent and scalable AI solutions
End-to-End ML Pipeline using Vertex AI
An end-to-end ML pipeline for predicting customer churn empowering businesses to proactively retain customers.
This project leverages VertexAI and Google Cloud for advanced machine learning capabilities and model versioning, and Streamlit for an intuitive user interface. It demonstrates the integration of cloud technologies with practical business applications in customer retention strategies.
Problem:
Customer churn is a significant issue for businesses, leading to revenue loss and decreased market share.
Solution:
Developed a machine learning pipeline to analyze customer data, predict potential churners, and provide insights through a user-friendly interface.
Insights:
The project likely enabled businesses to identify at-risk customers and take targeted retention actions, potentially improving customer retention rates.
Project Flow:
1. Data ingestion and preprocessing
2. Model development using VertexAI
3. User interface creation with Streamlit
4. Containerization with Docker
5. CI/CD pipeline automation
6. Model versioning and management
Stack:
Python, Docker, Google Cloud Platform (VertexAI, Cloud Storage, BigQuery, Cloud Build, Cloud Run), Streamlit
Automated Football Video Analysis
Computer vision-based system for automated analysis of football matches, providing detailed performance metrics.
Problem:
Manual analysis of football matches is time-consuming and may miss important details about player movements and team dynamics.
Solution:
Developed an automated system using computer vision techniques to track players, the ball, and referees, enabling comprehensive performance analysis.
Insights:
The system likely provides coaches and analysts with detailed data on player movements, team possession, and overall game dynamics, facilitating more informed strategic decisions.
Project Flow:
1. Object detection using YOLO
2. Player tracking with optical flow
3. Differentiation between player and camera movement
4. Calculation of player speed and distance covered
5. Analysis of team possession and performance metrics
Stack:
Python, OpenCV, Ultralytics (YOLO)
Image Segmentation with SAM2
Implementation of Meta's Segment Anything Model (SAM2) for advanced image segmentation tasks.
Problem:
Traditional image segmentation methods often lack flexibility and struggle with complex or ambiguous object boundaries.
Solution:
Implemented the Segment Anything Model (SAM2) to perform advanced image segmentation tasks, including automatic mask generation and interactive point-based segmentation.
Insights:
The project likely demonstrated the potential of SAM2 in various computer vision applications, possibly improving accuracy and flexibility in tasks such as object detection and image and video editing.
Project Flow:
1. Implementation of SAM2 using PyTorch and CUDA
2. Development of automatic mask generator for object segmentation
3. Creation of interactive point-based segmentation system
4. Exploration of potential applications in object detection and image editing
5. Performance evaluation and comparison with traditional methods
Stack:
Python, PyTorch, OpenCV, CUDA
.about
Based in Mumbai, India. I’m a Data Scientist, turning data into actionable insights with a focus on MLOps. Obsessed with sports stats and always chasing optimization, I believe data can solve anything—except maybe how many snacks I can devour after a game (still unmeasured). Let's connect if you are into AI, quirky ideas, and a touch of sports banter!
.say hello
why don't you come say hi
Email: mayur.parab1223@gmail.com
LinkedIn: linkedin.com/in/mayurparab23
GitHub: github.com/mayur-kun