Build a churn prediction model using boosting algorithms to do the following

1. End-to-end ML pipeline in Python

2. Deployment-ready API

3. Dashboard with churn probability and SHAP-based explanation

4. PowerPoint summary for business stakeholders

• Handle and process pipeline for extracting disaster response survey data from PostgreSQL Database
• Conduct prompt engineering exercise to obtain relevant outputs
• Supervised Fine-tuning of LLM where necessary
• Connect to FAISS Db for context development (RAG)

Autonomous Vehicle Trajectory Prediction - Developing machine learning models for accurate real-time vehicle behavior prediction to enhance autonomous driving systems in Europe.

• Developed vehicle trajectory prediction models using LSTM in Keras and Scikit-learn, improving prediction accuracy by 15% and enhancing simulation reliability.
• Integrated and optimized large-scale trajectory datasets, reducing training time by 25% and error rates by 18%.
• Conducted NLP-based information extraction from disaster response data using BERT and Llama 2 to summarize disaster summary.
• Designed Python-based pipelines for time-series processing, streamlining analysis workflows by 20%. 

Developed privacy-preserving systems as part of Bosch?s Mobility Innovation Co-creation Platform, designed to process and analyze large-scale traffic video data securely and efficiently.

• Designed and deployed advanced object detection models (YOLOv6) to automatically mask faces and license plates, ensuring compliance with global privacy regulations like GDPR.
• Trained the model from scratch on localized datasets to improve model precision by 30%, enhancing platform usability.
• Deployed a scalable, containerized application using Docker on AWS Sagemaker, enabling high-throughput processing of urban traffic feeds.
• Implemented robust privacy enhancement pipelines using Python and OpenCV, reducing manual intervention by 80% and ensuring consistent compliance across datasets.
  

Developed AI-driven modules for real-time traffic violation detection and visualization, empowering local authorities with actionable insights through Bosch?s Mobility Innovation Co-creation Platform.

• Built image recognition models using CNN-based deep learning techniques for accurate traffic violation detection, improving detection rates by 30%.
• Integrated real-time edge AI inference on devices, reducing system latency and enabling quick response times for authorities.
• Processed over 10,000 vehicle images using the IDD dataset to train models, resulting in a 21% improvement in vehicle identification accuracy.
• Developed interactive traffic analytics dashboards using Streamlit and Plotly, including heatmaps, trend predictions, and congestion insights, enabling city officials to make data-driven decisions.
• Automated deployment workflows using CI/CD pipelines (GitHub Actions), streamlining updates to live environments and reducing time-to-market.

Traffic Monitoring Dashboards - Developed real-time dashboards for traffic planners using city-wide camera data.

Analyzed GPS data from a bike-sharing service to improve user satisfaction and operational efficiency.

• Increased bike-sharing usage time by 18% through pattern recognition and optimized service allocations.
• Designed interactive dashboards using Plotly for real-time performance tracking.