Built a scalable, distributed data processing pipeline with PySpark and Azure ML, facilitating parallel machine learning model training on extensive stock market news datasets.Developed an interactive frontend with React.js, allowing users to submit news article links and interact with the Azure OpenAI-powered chatbot. Enhanced initial text classification and filtering through Azure Cognitive Services to optimize the chatbot’s performance on complex queries.Achieved a 50% increase in response accuracy by optimizing model usage with Azure Cognitive Services, reducing unnecessary large language model (LLM) calls through effective preprocessing.Improved the GPT-4-turbo model's relevance and accuracy by 20% in financial contexts through fine-tuning on domain-specific data.Reduced model training and inference times by 30% using GPU acceleration with Azure ML, enabling high- performance real-time computations.Successfully managed data scalability for up to 2,000 concurrent users by implementing efficient indexing and caching mechanisms with Azure Cosmos DB.Engineered a robust backend using FastAPI to handle API requests and integrate NLP services efficiently.Leveraged Docker and Azure Kubernetes Service (AKS) for deployment, implementing autoscaling to handle traffic surges cost-effectively, ensuring a consistent user experience.Led the deployment and scaling of services, ensuring seamless operations and cost management for both CPU-bound and GPU-bound tasksDemonstrated improved user engagement and decision-making for investors by delivering timely, accurate information to support stock market investment strategies.

Conducted Entity Classification on document images using the PyTorch toolbox Mmdetection, optimizing data mapping team productivity, resulting in an additional $650K in revenue and a 70% increase in customer acquisition.Employed GPU acceleration with CUDA to attain an 82% mean average precision in object detection, leveraging advanced algorithms like FasterRCNN, to enhance accuracy.Oversaw comprehensive tasks including data annotation, model training, REST API deployment, and seamless frontend integration, ensuring smooth operational workflows and optimal performance.Collaborated with a cross-functional team to customize the LabelImg annotation tool for data preparation. Managed a team of 8 members, overseeing annotated data verification and leveraging existing model inference to achieve a tenfold increase in annotation.Researched document classification methodologies, with a focus on leveraging pre-trained language models such as BERT and Transformer-based architectures. Investigated effectiveness across diverse document types to inform future project enhancements and optimize classification accuracy.Developed a data analysis Python script to preprocess image data for optimal modeling, including normalization, orientation adjustments, resizing, and pixel manipulation, merged this script seamlessly with a Flask API to ensure efficient data preprocessing and compatibility with downstream modeling tasks.

Supported to the development of AI models for various applications, focusing on Python-based backend development.Explored advanced machine learning (ML) algorithms and techniques, including deep learning and reinforcement learning.Enhanced data entry efficiency by 80% through effective table detection on images using the RetinaNet deep learning model.Achieved 95% accuracy in extracting text from table cells using Pytesseract OCR tool by detecting table grids. Employed computer vision techniques including dilation, Gaussian blurring, and binary thresholding to enhance text visibility in table images.Implemented a data augmentation pipeline using OpenCV and NumPy to generate diverse variations of image data, expanding the training dataset and improving the robustness of deep learning models.Attained a notable 78% F1 score in text classification through comprehensive statistical analysis employing Multinomial Logistic Regression. Implemented a project using python, skit-learn, NumPy, SciPy and UNIX for detecting handwritten digits using artificial neural networks (Natural Language Processing) and convolutional neural networks.Developed a database using JAVA, SQL, and UNIX for managing a library, with one user as a librarian with limited access and the other as a database administrator.Evaluated and optimized performance of models, tuned parameters with K-Fold Cross Validation.Provided analytical support to underwriting and pricing by preparing and analyzing data to be used in auctorial calculations.Designed dashboards with Tableau 9.2 and Meteor JS provided complex reports, including summaries, charts, and graphs to interpret findings to team and stakeholders.