Simulation, Sim Realism. We design the technologies that provide the most realistic simulations to evaluate the Waymo Driver.

(2023 - now)

• TL responsible for agent quality, integrating Waymo's latest foundational world model developments to create better sim solutions (simulation rollout, synthetic scenario creation).

• TL responsible for creating a unified pipeline to evaluate each onboard release, with a focus on using sampling-based approaches (generating dense counterfactuals) to compute trustworthy rates estimates of low likelihood events (e.g. collisions) with tight CIs.

(2021 - 2023)

• Modernized the agent stack from using hardcoded heuristics to model-based solutions, improving simulation realism by 20%

• Designed metrics and evaluation methodologies to evaluate sim realism.

Keywords: Foundational World Models, GenAI, Model Evaluation, Synthetic Scenario Generation, Sampling-based techniques

Payments Platform, Accounting. We serve over 60 products, and support 20B monthly transactions in 200 countries/regions and in 60 currencies.

(2018 - 2021)

• Created a unified workflow management system for the payments stack, and onboarded 3 partner teams onto the system.

• Redesigned accounting logic from hardcoded rules into a configuration-based system.

(2016-2018)

• Part of the initial 4-member team that rewrote the initial MapReduce-based accounting pipeline into an incremental, event-driven service. This reduced the duration of the month-close process from 7 days to about 4 hours.

Keywords: Distributed Systems, Database Design, Configuration-based Systems, Flume, Microservices

Part of Desktop Client. Designed a testing framework to run functional tests that verify the robustness of the desktop client, especially when new features are introduced.

Part of EC2 Security. Worked on Public Кey Infrastructure (PKI). Created a distributed metric reporter that periodically generates statistics regarding the status of certificates in hosts. Created the frontend website and backend functionality for a service that backfills certificates to hosts. AWS tools used include DynamoDB, Elastic MapReduce, S3, Data Pipelines, and Simple Queue Service.

Student Researcher at Stanford Infolab. Investigated the properties of different real-world networks and derived ways to classify them. Used machine learning to predict network type (social network, citation network etc) of a given graph. Added new functions into the Stanford SNAP library for analysing large networks. Currently working on generating synthetic networks that resemble real world networks.