Developed and optimized models such as CNNs, Vision Transformers, and ResNet Hybrids to classify jute pest images into 17 categories.
Achieved 98% accuracy through extensive experimentation, hyperparameter tuning, and performance evaluation.
Developed strategies to accelerate GPT-2 model inference using DeepSpeed and Distributed Data Parallel (DDP) frameworks on HPC systems, optimizing metrics such as throughput, latency, and resource utilization.
Conducted a comparative study across various configurations, demonstrating DeepSpeed’s efficiency in memory management for smaller setups and DDP’s superior scalability and performance for larger tasks.
Developed a website using Django and Python to give users personalized recommendations. Collaborative filtering and machine learning techniques are used to generate recommendations.
An intelligent expert system using Python to diagnose a mental disorder and recommend treatment of the diagnosed disorder. Forward chaining is used to diagnose the disorder and backward chaining is used to recommend the treatments.
A Website using Python, Django and D3 JS framework to help university authority to conduct various administrative works, see employee hierarchy in a nice graphical way and so on.