● Developed a solution for reducing redundant math questions in a database by 13%, leveraging advanced Natural Language Processing techniques using BERT and GPT-4 for similarity detection. Enhanced future scalability by integrating a MongoDB Vector Database.
● Engineered an end-to-end flow to reduce latency by 8% in image storage and retrieval. Implemented image compression and asynchronous task execution for efficient storage in Azure Blob, coupled with Redis Queue to manage task scheduling.
● Built and deployed an end-to-end system achieving 75% accuracy in mathematical data extraction. Utilized Python, Mathpix, and OpenAI for advanced Optical Character Recognition (OCR) techniques, and incorporated Redis caching for optimized retrieval performance.
● Fine-tuned a BERT-based model to classify 650,000 math questions into educational topics with 70% accuracy. Deployed the model on Hugging Face to enhance educational chatbot interactions and facilitate real-time question categorization.