Work

  • November 2024 - Present
    IBM
    AI Intern
    Developed a multi-stage AI pipeline to evaluate handwritten answer sheets using LLMs, achieving 88% higher accuracy than traditional OCR.Integrated IBM’s internal data ingestion software to convert PDFs into Parquet format, maintaining structure and layout for efficient processing.Benchmarked and optimized handwriting recognition by testing 10+ Vision-Language models, including WatsonX, Pixtral, Gemini, and other state-of-the-art models.Implemented LLaMA 3.2 90B Vision Instruct model, significantly improving answer extraction accuracy in complex answer formats.Designed a hybrid grading system using cosine similarity, sentence transformers, and adaptive NLP techniques, dynamically switching algorithms based on institutional grading preferences.Engineered an advanced diagram processing module, enabling AI to interpret and analyze handwritten diagrams with structured evaluation.Pioneered a custom regex-based chunking strategy, enhancing segmented answer evaluation and boosting efficiency in large-scale answer sheet processing.
  • March 2024 - September 2024
    Intel Technologies Pvt. Ltd.
    Project Intern
    Created multiple AI-powered internal tools to optimize data processing and model performance.Gained hands-on experience with LLM fine-tuning techniques.Explored Graph Retrieval-Augmented Generation (RAG) for enhanced data retrieval and understanding.Developed an internal tool to convert natural language to SQL using defog/sqlcoder-7b-2 from HuggingFace improving their old model improving the overall performance by 23%
  • January 2024 - May 2024
    Omdena
    Generative AI Engineer
    Led the LLM team of 30 engineers for a mental health support chatbot.Implemented RAG architecture to enhance chatbot performance.
  • February 2024 - May 2024
    Optimum AI
    AI Researcher
    Contributed to a hierarchical multi-agent AI finance planner project.Deployed solutions on AWS for scalability and performance.
  • September 2023 - 2024
    IEEE EPICS
    IEEE Research Intern
    Led a team of 20+ members in designing and developing a multi-terrain autonomous rover.Utilized LiDARs, sensors, edge computing, and machine learning technologies.