Harshika Pathak

Worked on Computer Vision projects such as object detection, OCR, and image classification. Built and used Generative AI models (open-source and closed-source like Mistral, GPT, Bedrock) for real-world tasks. Created RAG (Retrieval-Augmented Generation) pipelines to improve search and response accuracy. Developed APIs and services to combine AI models and make them production-ready. Collaborated with different teams to deliver AI solutions that match business needs. Researched and experimented with new AI techniques to improve performance and efficiency.

Fine-tuned object detection models such as YOLO and other deep learning-based detectors for document and field-level identification tasks. Trained and optimized LayoutLM-based models to extract structured key-value pairs from KYC documents and other scanned forms. Developed CNN-based solutions for visual pattern recognition and integrated Generative AI techniques for intelligent content generation and augmentation in fraud and document analysis workflows. Created and optimized prompts for LLMs to identify fraudulent insurance claims and check medical relevancy in healthcare documents and narratives. Deployed and integrated advanced Computer Vision (CV) and Generative AI models into production environments for scalable document processing solutions. Familiar with version control systems such as Git, Bitbucket, and GitHub. Collaborated closely with product managers and software engineers to align technical solutions with product requirements.