· Utilized advanced statistical models and machine learning algorithms (Pandas, Matplotlib, Scikit-learn) to identify trends and predict optimal purchase allocations, resulting in a 20% reduction in costs and improved sales performance.· Led the design and migration of over 2 million wholesale inventory records into a streamlined PostgreSQL database, improving data accessibility and query performance by 50%. Built a robust Python-Django-based API to manage inventory, sales, purchases, resource allocation, delivery routing and accounting processes in real-time.· Built an intuitive React-based management UI to automate daily operations. Deployed the application on AWS, ensuring high availability and scalability while managing the full project lifecycle using Git.

· Developed a highly responsive chatbot agent using ChatGPT Assistant API, fine-tuning it with industry-specific knowledge to answer remodeling-related queries. Continuously optimized its conversational abilities based on user feedback, improving response accuracy by 40%.· Designed and implemented UI components for a dynamic web app tailored to the remodeling industry, streamlining user flows for estimation tools and consultant proposals, which increased user engagement by 30%. Authored comprehensive requirements documentation for user registration and onboarding, ensuring seamless integration of new users.

· Loaded and preprocessed ECG time-series data from 4 distinct datasets consisting of over 100 subjects, utilizing PyTorch to detect R-peaks and segment signals for further analysis.· Developed a biometric authentication system leveraging convolutional neural networks (CNN) and transformer architectures, optimized using triplet loss and hyperparameter tuning to enhance model performance.· Achieved state-of-the-art results with a 4.19% Equal Error Rate and an 85ms processing time, outperforming existing approaches. Successfully reduced the False Reject Rate by 6%, demonstrating expertise in time-series data handling, deep learning, and model optimization for real-time applications.

· Competed in a lightweight optical Chinese character recognition challenge, where I compressed the model by pruning 90% of the FLOPS without compromising accuracy (99%), ranking in the top 15% of over 200 teams.· Engineered a supervised SRGAN model and unsupervised CycleGAN architecture for large-scale satellite image super-resolution and enhancement, optimizing performance through hyperparameter tuning, integrating perceptual, style, and adversarial losses, and achieving 30 PSNR, 0.8 SSIM, and superior visual fidelity on 10,000+ satellite images.· Developed an optimized non-linear activation function to minimize storage and computational overhead for FPGA-based neural networks, utilizing a novel two-fold Look-up Table architecture. This work led to a publication in an IEEE journal and demonstrated a 30% improvement in processing efficiency.