Experience

Meta - Software Engineer

San Francisco Bay Area, CA, January 2025 - PRESENT

  • Core engineer on Meta Ads Infrastructure, developing low‑latency C/C++ services powering global ad delivery.
  • Designed large-scale ML calibration systems that increased revenue by ~$10M/year and reduced fleet CPU usage by 2–3% across services handling millions of RPS.

Amazon Web Services - Software Engineer

San Francisco Bay Area, CA, January 2020 - January 2025

  • Database kernel (MySQL/Postgres) engineer on the AWS Aurora Database Engine team using C/C++.
  • Core engineer on the new Aurora Limitless product, presented at 2024 re:Invent (https://aws.amazon.com/about-aws/whats-new/2023/11/amazon-aurora-limitless-database/).
  • Led initiatives to create a sharded distributed storage engine capable of handling more than 1M TPS.

Givex - Software Engineer

Toronto, ON, January 2019 - January 2020

  • Backend development using Python, JavaScript, C/C++, PostgreSQL, and Unix.
  • Implemented gift card, point-of-sale, and other transaction-based solutions for Fortune 500 companies.

Projects

1st Place – Google Technology Prize, Hack for Social Impact 2025

Python, React Native, Node/Express, Google Cloud

  • Created a fully functional cross-platform app that acts as an AI assistant for youth facing homelessness.
  • Enabled users to chat with the AI assistant via text or speech; the assistant could also place phone calls to agencies on the user’s behalf without using the user’s phone number.

Two Sigma - AI Competition - Halite 2

Python, Tensorflow, Numpy

  • Designed and trained an RL-based AI bot to beat other bots in a virtual game created by Two Sigma (Ranked top 10% of players in Halite 2)
  • Scraped data from google cloud buckets and trained my bot on google cloud virtual machines

DattaBot – Artificial Intelligence (AI) LLM Bot

Python, PyTorch

  • Created a transformer LLM model, API, and bot. Uses latest PyTorch features, error and debugging logging, easy model hyperparameters and configurations, and much more.
  • MoE based model. Uses FSDP, GeLU, AMP, RMSNorm.