Arnab Sen

• TL, Capping AdBrain Model Predictions: improved the capping framework to constrain over-prediction using Prediction Intervals in a Poisson Log Loss DNN predicting conversion rates (pCVR); calibrated over-prediction profiles in highly sparse, privacy-restricted spaces like ITP, bringing Revenue Wt. Bias from +11 to −3.
• Lead, Attributed Branded Search (ABS) Modeling: leading the ML strategy and measurement for ABS within the Glitter project to recover conversion losses from signed-out traffic, targeting a $XXXM ARR impact.
• Project Lead, View-Through Conversion (VTC) Modeling: engineered and launched an end-to-end ML system to recover privacy-impacted VTCs for YouTube Demand Gen ads, projected to generate $XXM in ARR. Led a team (including an L3 engineer) and collaborated with Data Science on a TFX-based pCVR model with iterative calibration.
• Optimized a high-throughput Flume pipeline processing ~20 billion daily ad impressions, integrating reporting for YouTube's bidding systems.
• Engineering leadership: promoted to L4 in 1.5 years with an "Outstanding Impact" rating; C++ Readability mentor; mentored 5 Nooglers, hosted 2 STEP interns, and conducted ~40 technical interviews.

• Built a toolkit to generate dynamic metaverses from user input using OpenAI's GPT models.
• Designed a user-friendly UI for creating AI Brains and uploading information to populate and personalize a metaverse.
• Executed API integrations to connect components across the product.

• Integrated new protocol and server APIs to enable cryptocurrency exchange for users.
• Built swap functionality for Bitcoin, Ethereum, Polygon, Solana, and ERC20 tokens in device firmware.
• Designed and implemented swap-flow screens in the desktop application.

• Designed and built an email-based notification system for the Ads Resource Engineering team — evaluated Google's internal notification platforms, wrote the design doc with component overviews and workflow diagrams, and shipped a working integration.
• Implemented features, fixed bugs, and wrote unit tests for an internal Ads Planning tool.

• Improved User Statically Defined Tracing (USDT) support in Bitcoin Core by adding new tracepoints and scripts leveraging eBPF (PR #22902, PR #23907).
• Conducted a session on "Bitcoin Script" for an audience of 150.

B.Tech in Computer Science and Technology — CGPA 9.62

A sequence-to-sequence framework using CNNs and Transformers to automate classical cryptosystem identification, achieving 96.72% accuracy on a 130k ciphertext dataset.