• Implemented a robust real-time model utilizing Machine Learning algorithms to classify fraudulent activities, With a traffic of approximately 10K RPS serving with a maximum latency of 30ms • Created real-time pipelines for ingestion of Velocity features and Terminal features to Machine Learning models • Utilised Kubernetes and Kubeflow to develop and deploy machine learning pipelines. • Optimized Kubeflow components for cost efficiency and scalability.• Upgraded Kubeflow manifests in deployment from v1.6 to v1.8 and tested on UAT setup• Oversee the maintenance and development of Kubeflow Notebook Images, including but not limited to (Jupyter, Codeserver, Scipy, Tensorflow, and PyTorch), ensuring they are suitable for both production and pre-production environments.• Implement foolproof procedures for the creation, deployment, and monitoring of Kubeflow Notebook Images to guarantee their reliability and stability.• Collaborate with development teams to understand their requirements and integrate necessary tools and libraries into the notebook images.

- Worked on AEPS (Aadhaar Enabled Payment System) fraud models to enhance detection and prevention capabilities.- Built Docker images for automated Jenkins pipelines, streamlining the CI/CD process.Orchestrated workflows for data models using dbt-Dagster integration to ensure efficient pipeline execution.- Continuously optimized Docker images for performance, security, and scalability, staying updated with the latest technological advancements.- Established and enforced best practices and standards for image maintenance, versioning, and documentation.- Provided guidance and support to team members involved in image maintenance, fostering a collaborative and productive work environment.

Trained and fine-tuned multiple LLM's models from Data Engine for different companies.

• Development and Application of RLHFStrategies: I train advanced language models, working on projects for leading clients in the industry and focusing on the naturalness and effectiveness of chatbots.• Quality Control: I implement exhaustive practices to ensure accuracy and relevance in chatbot responses, enhancing the user experience.• Quality Assurance of Codes and Tasks: 1 ensure high quality and precision in all tasks and codes used in the training process, ensuring robust and reliable Al models.Integration of Human Feedback: | introduce human feedback into the model training cycle, adjusting and fine-tuning Al systems to improve empathy and context.

- Developed and implemented RLHF (Reinforcement Learning with Human Feedback) strategies, advancing language models and delivering state-of-the-art solutions for leading industry clients, with a focus on optimizing the naturalness and effectiveness of chatbots.- Enforced rigorous quality control measures to ensure the precision and relevance of chatbot responses, significantly improving the overall user experience.- Maintained excellence in code quality and task execution, fostering the development of robust and dependable AI models.- Implemented a comprehensive framework for integrating human feedback into the model training process, fine-tuning AI systems to enhance their empathy and contextual understanding.

As a software engineer, I spearheaded the development of "High Court Connect," a cutting-edge web application designed to empower lawyers and advocates in the digital era , By automating the information retrieval process, we save valuable time for legal professionals, enabling them to focus on strategic analysis and crafting compelling arguments.Product Perspective:1. High Court Connect is a web application designed for lawyers and advocates.2. Users can access the application via a web browser from a desktop or mobile device.3. The application accepts keywords from users to search for specific past judgments.4. It performs web scraping to find judgments related to the provided keywords.5. The retrieved data is processed and presented to the user through the web application.In summary, High Court Connect aims to simplify the information retrieval process for lawyers and advocates by automating the search for past judgments using web scraping and machine learning techniques. The application allows users to search for relevant judgments, filter the data, and access their search history.

1. Analysis & Implementation of “Predictive Analysis of Client Wise Sales” . 2. Predictive Analytics helps to analyse the trend of the data, evolve the risk factors associated and Preventive Actions.