Rohit Jain

Working on E2E recovery to allow one-click restore functionality for enterprise Postgres databases. Decreased PostgreSQL WAL log backup times from 10 hours to 5 minutes by decoupling backup and archival processes and implementing parallel ingestion, ensuring concurrency and consistency via shell scripting.

• Working on E2E recovery to allow one-click restore functionality for enterprise Postgres databases.
• Decreased PostgreSQL WAL log backup times from 10 hours to 5 minutes by decoupling backup and archival processes and implementing parallel ingestion, ensuring concurrency and consistency via shell scripting.

Led the redesign of metadata store replacing Cassandra with RocksDB and scaling support for deep storage with read-heavy workflows from 100 physical nodes to more than 300 nodes. Optimized the iSCSI target digest computation process by introducing a shared locking mechanism resulting in a remarkable decrease in login times and scaling simultaneous logins from 128 targets to 2000+ targets. Pioneered the implementation of a robust network interface hosting mechanism on multiple nodes, providing multi-path IO services to clients, dramatically reducing iSCSI reconnection latencies from 15s to under 1s. Won Nutanix hackathon 2023 for innovating a snapshot chain management solution over S3, reducing the primary cluster storage requirement to 1/10th saving crucial space for other metadata operations.

• Led the redesign of metadata store replacing Cassandra with RocksDB and scaling support for deep storage with read-heavy workflows from 100 physical nodes to more than 300 nodes.
• Optimized the iSCSI target digest computation process by introducing a shared locking mechanism resulting in a remarkable decrease in login times and scaling simultaneous logins from 128 targets to 2000+ targets.
• Pioneered the implementation of a robust network interface hosting mechanism on multiple nodes, providing multi-path IO services to clients, dramatically reducing iSCSI reconnection latencies from 15s to under 1s.
• Won Nutanix hackathon 2023 for innovating a snapshot chain management solution over S3, reducing the primary cluster storage requirement to 1/10th saving crucial space for other metadata operations.

Developed an error injection framework for failing gRPC requests reducing simulation times by 80%. Extended the framework for detection of slow nodes with minimal false positives and latency of less than 100ms.

• Developed an error injection framework for failing gRPC requests reducing simulation times by 80%.
• Extended the framework for detection of slow nodes with minimal false positives and latency of less than 100ms.

Created an android application that learns the user color-tone preferences via on-device learning approach. Trained a GAN-based Model for user behavior learning with split cloud/on-device approach, achieving score of 90+.

• Created an android application that learns the user color-tone preferences via on-device learning approach.
• Trained a GAN-based Model for user behavior learning with split cloud/on-device approach, achieving score of 90+.

Worked on entropy-based model for unifying categorical data to perform clustering for 10000+ data rows. Used divergence model for classification of healthy individuals and achieved an accuracy rate of over 95%.

• Worked on entropy-based model for unifying categorical data to perform clustering for 10000+ data rows.
• Used divergence model for classification of healthy individuals and achieved an accuracy rate of over 95%.

Selected into first cohort at Perplexity AI Business Fellowship 2025.

Patent for improving temporary write buffer (oplog) flushing, achieving eightfold performance gains by optimizing drain speeds for consistent I/O performance during various states (e.g., idle, rebuild) when serviced by the persistent storage tier or SSD-based oplog.

Published a PyPi package for finding a common distance matrix for ordinal and nominal data using entropy measures.