Shini Mary Koshy

As part of WatsonX product engineering team, we leverage the WatsonX platform to co-create AI value with clients, focusing on technology patterns to enhance repeatability and delight clients. Working as innovative AI Solution Engineer ,specializing in foundation models and large language models, responsible for architecting and delivering AI solutions using cutting-edge technologies, with a strong focus on foundation models and large language models.

Project done so far:

1.Metro - Marketing task

• Implemented LLM-based tools to analyze consumer sentiment, optimize ad copy, and enhance customer engagement. Automated content creation for Metro's digital marketing campaigns, leading to an increase in engagement.Developed AI-powered chatbots to improve customer interaction and reduce response time.

2.Juki- Augusberg:

• Developed and implemented Juki, an AI-driven assistant to streamline child welfare case management.Automated documentation, case tracking, and reporting, reducing administrative workload.Enhanced decision-making for social workers by providing AI-based insights and recommendations.

3.Agent for Hensoldt INCOSE Guideline checking

• Designed an AI-powered agent to automate compliance verification with INCOSE (International Council on Systems Engineering) guidelines for Hensoldt.Check each requirements and validate with agents whether they follow INCOSE Guidelines

• Led a team of data extraction specialists and OCR experts in document processing projects
• Developed and implemented a HYBRID OCR solution for enhanced PDF document extraction using pdfminer and Tesseract, resulting in increased efficiency
• Trained and fine-tuned PADDLE OCR and TROCR models for handwritten recognition and created custom trainable OCR solutions
• Collaborated with cross-functional teams to improve data extraction processes and ensure accuracy
• Conducted quality assurance checks and implemented best practices to optimize OCR performance
• Utilized Tesseract, Azure OCR, and Amazon Textract for data extraction from invoices, ID cards, and forms
• Ensured accurate conversion of printed and handwritten content into digital formats
• Conducted OCR model training and customization for specific document types
• Leveraged GENERATIVE AI techniques to develop a Natural Language to SQL query generation system
• Fine-tuned large language models (LLM) with custom datasets to generate organization-specific queries, integrating Azure Open AI GPT and Hugging Face Transformers
• Collaborated with clients, customized AI solutions, and drove innovation in generative AI and NLP through continuous learning and training

• Project Overview: Apple Account
• Conducted extensive data analysis on a customer dataset containing years of transaction history
• Cleaned and processed the data, addressing missing values and outliers
• Worked on analysing the data and getting insights from the data collected
• Provided critical insights to the business team, enabling data-driven decision-making processes
• Analysis served as a foundation for strategic choices and initiatives within the organization

• Project Overview: Cannon Medicals
• Detection of heart cycle peaks using YOLO V3 model: The model is trained and integrated into ultrasound machine for detecting E and A points in Heart Cycle
• Integrated the model into ultrasound equipment, enhancing diagnostic capabilities for cardiac conditions
• Detection of Edges on Carotid Veins for Arthritis Diagnosis: Created a custom CNN for precise carotid vein edge detection, aiding early arthritis diagnosis, and enhancing patient care through early intervention
• Segmentation of Arthritis in Ultrasound Images: Built a robust deep learning model for accurate arthritis region segmentation in medical images, enhancing disease analysis, staging, and treatment planning for medical professionals
• Image Denoising using GAN and DNCNN: Applied GANs to enhance DNCNN's image denoising by generating random noise, significantly improving image quality, particularly in noisy camera images