Venkata Narayana Redrouthu

In my capacity as a Research Assistant at the Stony Brook Cancer Center , my work has centered on computational biology and single-cell multiomics analysis. I successfully migrated a complex single-cell multiome (RNA+ATAC) analysis pipeline from R to Python, utilizing Scanpy, Snapatac2, and AnnData to significantly enhance both processing efficiency and reproducibility.

DS-AttentionUNet: High-Recall Skin Lesion Segmentation — Python, PyTorch, OpenCV 

• Designed a novel DS-AttentionUNet architecture in PyTorch with Attention Gates and Deep Supervision to improve model convergence for medical image segmentation.
• Achieved a high Recall of 92.16% by implementing a sophisticated training pipeline with a composite loss function, mixed-precision training, and a OneCycleLR scheduler.
• Assembled a modular, end-to-end application with a command-line interface for streamlined training, prediction, and inference.

Automated Document Classification with LLMs — Python, LLMs, OpenSearch, Django  

• Conceptualized a classification framework processing 100K+ documents, achieving 73% accuracy with a custom LLM-based hierarchical taxonomy.
• Optimized document retrieval latency by 40% by implementing OpenSearch and custom vector indexing for over 10K+ queries per hour.
• Developed a Django app that cut visualization time from 4 hours to 5 minutes.

Development of Custom R Statistical Package — R, R Markdown 

• Architected a production-grade R package with 15+ statistical functions, improving learning outcomes by 50% for 50+ students.
• Reduced the learning curve by 60% by creating comprehensive documentation and tutorials using R Markdown.
• Instituted a robust error-handling system that reduced runtime errors by 80%.

Predictive Risk Modeling for House Price Forecasting — Python, Scikit-Learn  

• Developed an ML pipeline achieving an 86.99% variance score using ensemble methods and advanced feature engineering.
• Optimized model performance by 25% through automated hyperparameter tuning.
• Automated a preprocessing pipeline for 100K+ records, reducing training time by 35%

• House Price Prediction Using Machine Learning Algorithms, International Conference on Emerging Trends in Electronics and Communication Engineering, Nandyal, 2023-04-01
• Big Data Analytics on Social Networks to Detect Melancholy in Real-Time, International Journal of Scientific Research in Engineering and Management, 06, 12, 2022-12-01
• Artificial Intelligence Chatbot Using Python, Journal of Engineering, Computing and Architecture, 12, 3, 2022-03-01
• A Smart Mirror Using Raspberry Pi Based on IOT, International Research Journal of Engineering and Technology, 08, 12, 2021-10-01