Passion Fuels Purpose! 

• Developed voice-enabled financial advisory platform using Nous Hermes 2 LLM and FAISS vector store, creating expense tracking, prediction visualization, and personalized Q&A features through RAG pipeline

• Enhanced G.A.I.N framework by implementing DCGAN architecture for missing data imputation, improving upon traditional GAN approach with 13% higher accuracy in data reconstruction

• Innovated silicon photonic component design by replacing VGGNet/ResNet with Data-Efficient Image Transformer (DeiT), achieving 12% better component generation through improved feature extraction

• Built university-wide research platform using GPT-4 and AWS Bedrock APIs, enabling semantic paper search and automated knowledge extraction for internal research data accessibility

• Architected end-to-end RFQ processing pipeline, integrating custom spaCy NER and ensemble models(AdaBoost and Random Forest) with containerized deployment using Kubernetes, achieving 95% accuracy in PLC price predictions

• Engineered full-stack solution for RFQ price prediction for Schneider Electric, utilizing a React.js frontend and FastAPI microservices with Jenkins CI/CD pipeline on AWS EC2, reducing manual quotation processing time from 7 days to 4 hours

• Prototyped ML pipeline for casting defect detection using YOLOv5 on AWS SageMaker in a 2-member team, demonstrating proof-of-concept with React.js dashboard and IoT integration achieving 78% detection accuracy

• Optimized existing P&ID extraction pipeline by fine-tuning Mask R-CNN model, achieving 6% accuracy improvement over Detectron 2 baseline

• Automated multi-threaded web scraping system using Python and Selenium, integrating with SQL Server for efficient storage and retrieval enabling concurrent data extraction from 25+ e-commerce platforms, reducing manual effort by 96%

• Built scalable PySpark ETL pipeline to process 10,000+ daily product records from SQL Server, automating segment-wise market analysis for South African e-commerce platforms

• Designed ML-driven customer segmentation model using K-means clustering on 5,000+ profiles, generating insights through Power BI dashboards that increased ad campaign’s CTR by 15%

• Implemented Google's Neural Style Transfer research paper, showcasing the fusion of artistic styles with images

• Researched advanced image classification techniques capable of distinguishing between over 1000 classes

• Authored findings from in-depth research in Neural Machine Translation and Deep Local Features on Medium`

• Directed Flutter-based indoor navigation system implementing A* pathfinding algorithm and custom geolocation services, reducing average path discovery time by 40%

• Deployed backend infrastructure using Python and MongoDB, integrating RESTful APIs for real-time location updates and path optimization with GCP deployment

• Architected CI/CD pipelines for automated testing and deployment, implementing containerization with Docker and establishing automated build processes on GCP