- Developing a Python chunking module for efficient data processing, enabling logical chunking of large documents while maintaining token limits and ensuring meaningful, sentiment-aware responses to queries.

- Building an evaluation framework to assess model outputs across key metrics including faithfulness, relevance, coherence, completeness, and answerability.

- Implementing topic modeling techniques for analyzing query topics and validating their coverage within source documents.

- Researching, implementing, and comparing various clustering algorithms to group similar queries, enhancing query understanding and improving system-level performance.

- Analysed and optimised campaign data: Conducted comprehensive analysis of marketing campaign performance metrics, utilizing statistical methods to identify trends in consumer behavior. Developed and implemented data-driven strategies to enhance campaign effectiveness, resulting in improved key performance indicators.

- Worked on campaign optimisation model based on Deep Q learning: Contributed to the development of an advanced machine learning model for campaign optimization using Deep Q learning techniques. Assisted in data preprocessing, feature engineering, and model training processes. Helped interpret model outputs to inform practical campaign strategies.

- Worked on Geospatial data to analyse Point of Interest and extract brands: Utilized GIS tools and spatial analysis techniques to process geospatial data, developing algorithms to identify and categorize Points of Interest. Implemented natural language processing methods to extract brand information from location data, contributing to the development of location-based marketing strategies.

- Collaborated with Prof. Manisha Verma on a project focused on human yoga posture detection: Led research efforts in computer vision and pose estimation techniques, specifically tailored for yoga postures. Worked closely with the professor to define project goals, methodology, and evaluation metrics. Contributed to the development of a novel approach for accurate and efficient yoga pose recognition.

- Utilized Graph Neural Networks (GNN) for the detection process: Implemented and optimized Graph Neural Network architectures for pose detection, leveraging their ability to model complex relationships between body joints. Conducted experiments to compare GNN performance against traditional convolutional neural networks, demonstrating improved accuracy in pose estimation tasks.

- Worked with a dataset containing images of various yoga postures. Focused on identifying joint points as node points in the network: Curated and preprocessed a comprehensive dataset of yoga pose images, ensuring diversity in postures and practitioners. Developed algorithms to accurately identify and label key joint points in each image, creating a robust foundation for the GNN model. Implemented data augmentation techniques to enhance model generalization across different body types and yoga styles.