Google

• Improved the quality of SmartLabel (Gmail's inbox sorter) training data, as measured by reduced human-evaluator bias, by designing and implementing a Python pipeline for cross-evaluation.
• Ensured user privacy and data security by removing Personally Identifiable Information (PII) from donated emails before cross-evaluation in the pipeline.
• Enabled business-critical insights into consumer behavior, guiding the development of an upcoming Gmail product, by building a distributed C++ pipeline that uses LLMs to securely analyze outgoing emails.

Cerebras Systems

• Migrated compiler backend to LLVM, enabling separate compilation of Cerebras-Assembly and C++ source files, improving ML kernel development flexibility and simplifying support for future hardware generations.
• Wrote algorithm to compress TableGen files, cutting size by 26%, line count by 18%, and generation time by 12%.
• Implemented scoping, register allocation and other language features using advanced data structures & algorithms.
• Developed a large volume of high-performance C++, learned about Cerebras' weight-streaming paradigm and utilization of MLIR & SSA to optimize graph-compilation of entire ML models to their wafer-scale engine (SoC).

Apple

• Designed and implemented a DMA controller driver feature (in C++) which streamlines PIO register programming and unlocks new avenues for enhanced performance through wider feature adoption.
• This enhancement reduced lines of code by 95% for each application, significantly improving feature usability.
• Debugged intricate Hardware/Software interactions involving IO-MMU and embedded co-processors on the SoC.
• Identified and resolved an existing memory allocation bug, reducing feature memory usage by 62.5%.

Google

• Enhanced the team's benchmarking framework (in Golang) with storage, visualization, analysis, and alerting features, ensuring the detection of performance regressions and boosting the rollout reliability of our product.
• Authored a 12-page design document outlining major decisions and future extensions to the project.
• Developed a library to improve scalability, resulting in a 90% reduction in test configuration time and lines of code.
• Reduced implementation time by 50% (4 weeks) by researching, proposing and adopting alternative internal tools.

Royal Bank of Canada

• Isolated the batch-running component from RBC's client onboarding platform, resulting in 10x faster build times.
• Upgraded the batch-runner's functionality using Java's Spring Boot Framework and deployed it to Cloud Foundry via an automated Jenkins pipeline, enabling non-real-time activity refreshes for over 1,000,000 clients daily.
• Overhauled a major Angular component to make it more reusable, reducing its content update time by 80%.

Languages and Databases

Python

C/C++

Java

Golang

JavaScript

TypeScript

PostgreSQL

Shell Scripting

Frameworks and Libraries

React

Django

PyTorch

Flask

Node.js

Spring Boot

Tools and Technologies

Git

Jenkins

Docker

AWS

FPGA

Arduino