Ai Developer

Led a 4-member R&D team to develop AI-driven solutions spanning LLMs, computer vision, and document analysis. Personally developed a Text-to-SQL system using open-source LLMs, achieving over 80% accuracy on real-world queries. Built an information extraction pipeline for medical billing documents using LLaMA 3.2 Vision, Pixtral, Mistral Small and OCR, reaching 95%+ precision through iterative testing. Proposed and implemented a new schema design for MS SQL Server to support LLM-driven data ingestion, enhancing maintainability and reducing data redundancy by 25%. Trained YOLOv8 for document section detection, boosting preprocessing accuracy by 40%. Used Git for version control and managed GPU clusters for efficient model training and optimization. Provided technical guidance and hands-on support throughout all stages of development. Distributed Mistral Small shards across GPUs, enabling multi-GPU inference and saving ~50% GPU memory per node.

Led the development of an AI-powered Text2Video tool using GPT-4, Stable Diffusion, and DALL-E, slashing content creation time by over 50%. Designed an OCR-based PDF to CSV conversion tool using textract, integrated with a Flask backend and served via cloud endpoint to boost admin efficiency by 75%. Developed an AI-based Exam Checker that automates test paper evaluation using OCR + LLaMA 3; deployed the system on Vertex AI, improving grading efficiency by 70%. Collaborated with cross-functional teams using JIRA to effectively address and resolve technical challenges, enhancing innovation in educational technology. Conducted research on emerging AI/ML techniques to enhance the functionality of educational tools continuously. Demonstrated expertise in developing scalable Flask web applications that support AI-driven educational platforms, thereby improving user experience.

Created an application to fetch necessary research papers from a database using LLM model GPT-4 along with BERT. Integrated LLM model(GPT-4), along with LangChain and semantic search to develop advanced, accurate chatbots. Engineered custom AI models, leveraging XGBoost and linear regression for predictive analysis. Led data preparation and model optimization, enhancing performance and outcomes. Evaluating models and iteratively optimizing for superior outcomes. Engaging with cross-functional teams, translating findings, and incorporating feedback.

Developed a facial emotion recognition system with 94.2% accuracy using a fine-tuned VGG16 CNN model. Sign language detection system using VGG16, achieving 92.5% accuracy on a custom dataset of hand gestures for non-verbal communication. Created a license plate recognition pipeline using YOLOv5 for detection and Tesseract OCR, achieving 88.3% accuracy. Implemented a real-time bike helmet detection system with YOLOv5s, reaching 87% [email protected] on a custom dataset of 4,000 images. Designed a student grade prediction model using Random Forest, achieving 89.6% accuracy on academic data.