Projects

Conducted a 42-run ablation study on DeBERTa-v3 using QLoRA, retaining 97% accuracy while reducing VRAM by 90%. Built a Multi-Agent Research Assistant using LangGraph to orchestrate agents for search, ranking, and summarization.

Engineered an end-to-end pipeline combining BLIP (Image Captioning) and FLAN-T5 (Story Generation). Built an automated product mockup system using Stable Diffusion (img2img), optimizing denoising strength for photorealistic textures.

Fine-tuned YOLOv8s on a custom COCO subset, managing the full lifecycle from JSON-to-TXT formatting to training. Benchmarked Generative models (GANs vs VAEs) on MNIST/CIFAR-10, proving GANs achieved superior image fidelity.

Developed a robust biometric security system to detect face spoofing (replay/print attacks) using Transformer architectures. Fine-tuned Swin Transformer achieving 99.75% test accuracy on CelebA-Spoof.

Engineered a multi-class classification model (Glioma, Meningioma, Pituitary) achieving 99.65% accuracy. Implemented Grad-CAM and Attention Rollout to visualize decision heatmaps.

Engineered a robust flood forecasting pipeline using CatBoost and XGBoost, trained on a massive dataset of 1.1 million+ records. Achieved a high predictive accuracy of 0.85 R².

Implemented a Transformer-based deep learning architecture to classify complex biological sequences, leveraging BERT embeddings for feature extraction.

Architected and deployed 100+ automation bots to handle complex workflows (YouTube engagement, file management). Integrated Computer Vision-based selectors for mobile testing bots.

Conducted time-series analysis on historical stock data to detect market manipulation patterns. Utilized Python (NumPy, Pandas, Seaborn) for data preprocessing, visualization, and predictive modeling.

Developed machine learning models to predict EV charging demand based on spatial and temporal factors. Generated insights for optimal charging station placement and smart grid load balancing.

Engineered an IoT-powered solar inverter with real-time Maximum Power Point Tracking (MPPT). Integrated a solar tracking system to enhance energy efficiency and power generation.

Developed robotic process automation (RPA) bots using WinAutomation and MacroDroid for YouTube engagement automation. Automated VPN switching, browser interactions, and video monitoring to enhance watch hours.

Developed and deployed WordPress websites on cloud platforms (AWS, Azure, DigitalOcean) using LAMP/LEMP stacks. Optimized hosting environments with cPanel, CyberPanel, and cloud-based caching solutions.

Optimized website speed and responsiveness, improving performance by 10% through caching, compression, and database tuning. Conducted technical SEO audits and implemented Google Search Console fixes.