• CMRG (Athletic Data Analysis)

1. Analyzed Functional Movement Screen (FMS) data using PCA and GMM to identify injury-prone movements, achieving 90% sensitivity and 86% specificity in females, 88% sensitivity and 100% specificity in males.

• 4D Cell (AI Drug Discovery)

1. Designed a contrastive learning pipeline (SimCLR, BYOL) coupled with VAE to compress microscopic images to identify mitochondrial phenotypes in cancer cells, achieving 75.8% accuracy across 26 drugs.

2. Achieved 20x speed up using GPU parallelization on legacy C++ mitochondrial analysis algorithm mitograph.

3. Developing 4D cell segmentation techniques using SAM and SAM2, and a point-cloud-based tracking algorithm to stitch 3D segmentations over time.

Developed a query analytics engine using KDB/Q to optimize prime brokerage resource allocation, reducing processing time by 2+ hours daily through efficient transaction tracking and capacity analysis.

Worked on Flipkart’s Adtech Platform team handling data and infrastructure challenges:

1. Druid Infrastructure Ownership:

• Led the migration of Druid to GCP, integrating Dataproc for ingestion, reducing processing costs by 15%, improving throughput by 36%, and scaling for peak sales.

• Architected a microservice for Ads ETL pipelines, enabling efficient batch and real-time processing, allowing clients to interact solely through a configuration service.

• Optimized SQL queries by transitioning to Druid-native queries, improving resource efficiency and reducing query latency.

2. Designed a fault-tolerant ML feature store using Druid and Kafka, enabling real-time feature updates for ranking models and CRM systems with low-latency access.

3. Engineered a funneling solution using Hadoop MapReduce to introduce cross-brand ads, reducing excessive query joins, and boosting search ad click-through rates (CTR) by 3-4% and fill rate by 8-10%

4. Integrated real-time keyword targeting using Spark, feeding data into a transformer model to improve product ad relevancy and boost fill rate by 4%.

5. Resolved cross-DC call issues in Apache Storm for real-time ingestion by implementing a communication abstraction, enabling traffic splitting across two data centers. Monitored live data center additions to debug real-time issues, achieving zero breakouts.

6. Scaled distributed ingestion pipelines on GCP to handle 6x traffic(2M RPS, 2 Petabytes per day) during sale events, leveraging Redis caching, Ansible for horizontal scaling, and automated load testing with Locust

Worked as a core member of the Strata TCAM Infrastructure team at Arista Networks:

1. Led the design and implementation of event-driven broadcast filtering in hardware, addressing producer-consumer mismatches and restart edge cases. Integrated CLI commands for client verification and compliance, ensuring seamless adoption and system efficiency.

2. Enhanced TCAM statistics export using TACC by improving design patterns and adding Python tests, increasing maintainability and security.

3. Implemented an SDK-level API to optimize in-memory operations, achieving low latency and faster rule processing in TCAM.

4. Debugged system-level networking issues with GDB and TCAM statistics, resolving broadcast/unicast faults, and segmentation errors, and ensuring system stability.