Building infrastructure to support LLM use cases across Google Photos, particularly for life understanding and narration.

--

Leading design and execution across the Memory Recommendation & Delivery space, collaborating cross team to optimize relevance and personalization of content, with a projected meaningful Daily Active User growth of 12%.

Maintaining a recommendation & notification server with a combination of Integer Programming and ML-driven signals using Java, C++ scaffolding, Boq & Spanner which serves 1B+ users.

Collaborating with Google Research and Google One, influencing design and decisions x-Product through tech talks, design reviews and data analysis (in Google SQL).

Designed and built infrastructure for notification recommendation and delivery with a server QPS (Queries Per Second) of over 100k using Java, C++ & Spanner, while collaborating with Google Research to use ML models to improve content curation, timing, and delivery.

Implementing infrastructure to switch from offline to online processing of car logs, in order to create a real-time world state in the autonomous vehicle itself, reducing processing time of these logs by over a day.

Implemented data pipelines for Action Moments and Photobooth ML models to pick interesting clips, resulting in 16% increased user coverage and engagement in early demos.

Built infrastructure for Suggestor nodes from ground up using C++ to filter out the interesting clips, and designed new RPCs in Java to pass this information to the web and mobile clients.

Consolidated the user data generation and profile pipeline by adding a new, automatically invoked testing phase using C++ and Spanner which reduced redundant tasks and decreased processing time from 2 hours to 45 minutes.