Primary architect and developer of a low-level, multi-threaded, high-performance C++ networking layer for Nuro’s real time remote driving platform. This layer enabled low latency + low loss real-time video streaming over variable network conditions.

Designed and implemented a novel congestion control algorithm using adaptive application-limited probing, early indicators (OWRD), and modem based forward error correction (FEC) resulted in 33x improvement in latency metrics.

Developed key components of video streaming stack:

sender side: packetization, multi-modem dynamic distribution algorithms

receiver side: jitter buffer and frame organizer for frame pacing/synchronization

Designed and implemented dynamic packet re-routing, using Linux TC and CGroup utilities.

Integrated custom C++/GStreamer QP estimation video codec wrapper to improve video quality (3x efficiency increase, 25% QP reduction).

Led extensive cross-functional data driven initiatives across four teams to address and resolve critical networking issues blocking company deployments.

Coordinated complex feature development across multiple teams, prioritized tasks

and leveraged collected network metrics for investigations and bug fixes.

Developed and implemented comprehensive benchmarking and profiling (perf, VTune) in C++ for Nuro’s model interpreter to enable CPU and memory optimizations of machine learning inference.

Built data pipelines to ingest critical network metrics into BigQuery.

Managed and analyzed large datasets, employing data cleaning techniques and

regression analysis to identify and predict network trends and patterns.

Worked a cloud platform to manage and schedule mapping missions

Built a mission scheduler to optimize mapping mission schedules based on distance

Implemented Go-based micro-services deployed in GKE which utilized various cloud technologies: MySQL, gRPC, Pub/Sub (GCP message queue)

Developed and built microservices with AWS technologies in Python for GE Health Cloud DataLake.

Led the design, implementation and deployment of a distributed virus scanning service with 300x the load capacity of previous system

• Multithreaded implementation had over 3x increase over original design

• Designed distributed synchronization for shared workload, cleanup, machine failovers

• Incorporated autoscaling for cost optimization and dynamic load handling

Led performance optimizations across platform:

• Added multithreaded configurations for SDK uploads resulting in 2x speedup

• Implemented buffered streaming for 25x decrease in memory usage during DICOM image processing

Utilized AWS technologies such as API gateway Lambda, EC2, Athena, DynamoDB, S3, Elastic Search, Cloud Formation, Cloud Watch, IAM, SQS, SNS.

Designed and implemented restful APIs and their corresponding lambda services. Also created unit, component and end to end tests for lambdas and APIs using mock, boto3 and behave.

The Edison Engineering Development Program seeks to create the next generation of engineers by combining 4 technical rotations (6 month long projects in all areas of the company) along with challenging coursework (mechanical, electrical and software).

First Rotation:

Digital Mammography:

Used Flask, Python and web development languages (HTML, JS, CSS) to create a data visualization tool for the image quality team.

Magnetic Resonance:

Developing APIs for the MR image reconstruction SDK.

Optimizing image processing algorithms for use with GPU.

GE Healthcloud:

Part of team leading the architecture and development of the next generation Data Lake for Healthcloud.

• Improved performance of Centricity Universal Viewer by implementing database caching and multithreading operations.

• Reduced image loading time by 25% and conducted performance testing.

• Built proofs of concept (image fetching servlet and search engine) for product troubleshooting improvement.