Prasad Pawar

Developed Surveillance for Capital Market and Anomaly Detection solution to identify abuse scenarios in the capital market. Worked on identification and prediction of suspicious entities using classification algorithms, model training, testing, evaluation techniques & optimization. Segmented the dataset based on business logic using Louvain community partition and K-means Clustering algorithms and represented it in the form of network graphs for better analysis. Developed Big data application from scratch implementing business logic using Python, Pyspark, and Pandas, deployed on Kubernetes clusters which helped in time optimized results by fine-tuning spark configurations. Applied various Machine learning techniques such as Topic Modeling, Sentiment & Summarization using BERT, Name Entity Recognition on communication data to get more understanding and relevance of fraudulent activities performed by participants of the capital market. Delivered solution for life science-based domain, "Early detection of cross binders in the drug discovery process" by applying Machine Learning and Deep Learning approach using neural networks, Random Forest, XGBoost algorithms for training and prediction. Developed a memory recommender system for High-Performance Computation applications to increase the availability of resources with a better estimation of memory before execution of the application on HPC clusters to achieve optimum utilization of HPC resources. Optimized OpenFOAM on Intel Xeon-Phi (KNL) by identifying hotspots using Vtune and applying AVX-512 intrinsics. Applied vectorization using SIMD pragmas to enhance the performance. Parallelized and optimized the IRS computation using CUDA-C on Nvidia K20, and using OpenMP and ICC on Intel Xeon Haswell-EP.

Improved performance of various image processing (ADAS) applications by code optimizations and parallelizing the source code using C, OpenMP, OpenCL, GPGPU, and Linux OS. Designed and implemented auto-parallelization of loops using YUCCA, and redesigned the automatic parallelization module as a Tech Lead. Worked on a project of performance enhancement and did various experiments on OpenMP constructs.

Designed and implemented a Disaster recovery module to achieve zero Recovery Point Objective (RPO) and negligible Recovery Time Objective (RTO), including automatic replication of data from DC site to DR site at block level using iSCSI protocol with PITR techniques of PostgreSQL. Patent granted for Method and System for Business Continuity and Disaster Recovery. Published research on Automatic Sequential to Parallel code conversion (S2P tool) and Enterprise Storage Architecture for Optimal Business Continuity.