Led engineering efforts across Amazon’s Marketing Measurement organization, building scalable systems that power data-driven campaign decisions across 10+ business units and global channels. Worked on high-impact cross-company initiatives, data orchestration infrastructure, and ML automation platforms to modernize how Amazon measures and optimizes marketing effectiveness.

Key Highlights:

* MMI Orchestration: Owned the Data Processing Service (DPS) of Amazon’s Marketing Measurement Infrastructure — a dependency-aware orchestration layer that triggers large-scale transformations based on Global Data Schema (GDS) metadata. Reduced processing latency from 8 weeks to 2 weeks and delivered model-ready datasets across 7 channels in 10+ countries.

* Meta/Amazon RCT Integration: Architected a cross-company system with Meta to run user-level randomized control trials (RCTs), enabling accurate lift measurement for $34M+ in social advertising spend using AWS Clean Rooms and privacy-safe matching.

* Geo-RCT Automation: Automated marketing measurement pipelines for geo-based RCTs and predictive models, cutting end-to-end latency by 80% and increasing measurement throughput by 4×.

* Self-Serve Campaign Platform: Built a platform for internal stakeholders to launch fixed marketing campaigns and measure lift autonomously — reducing reliance on applied scientists and saving ~90% of manual effort.

* ML-Powered Ad Mapping: Delivered a creative fingerprinting and video matching system that eliminated ~150 hours/year of manual mapping work through ML-driven automation.

Lead the design and development of a self-serving Risk Control platform. It provides the framework to onboard risk control using JSON Schema, place/lift user restrictions, send proactive user communication, and collect user activities using Kafka pipeline for auditing. Further, it assists to perform user impact analysis using shadow mode testing, and Looker dashboard for analytics. This helped mitigate the R0 risks, avoid millions of dollars in reparation, and made Robinhood compliant with SEC regulations.

Server Side Tracking

With the latest privacy changes to iOS, chrome, and other platforms, the industry is moving from pixel tracking to sending the site events to marketing partners from server side. This process is known as Server-Side Tracking

I have worked on creating a real-time application/framework to enable Server side tracking for all the marketing partners like google, Facebook, criteo, and so on. I lead the entire SDLC of this project starting from requirement gathering, design to development, deployment, and validation.

This golang application reads data from Kafka, processes it in the desired format, and then sends it to marketing partners. It is highly robust, self-healing by using a circuit breaker and optimized with the ability to process around 150,000 messages/sec with just 2 ec2 machines.

This project helped us with improving fidelity, security, user privacy and giving the marketing team more control over the data being shared with marketing partners compared to using pixel. This also removed the need for loading pixel on the site and thus improving the site speed.

Product Price data for Marketing Feeds:

To show the new and existing products on the ad platforms and marketplaces, we generate marketing feeds that consist of product information. One of the critical parts of this process is to keep feed price data in sync with the site prices with applied promotions and mark-downs; which otherwise leads to feed rejections from marketing partners.

I collaborated on creating a gRPC Java client application that fetches the product pricing data using the Server Streaming rpc endpoint from the pricing service. Once the data is in-memory, it is uploaded to S3 using Producer-Consumer multi-thread architecture. This application works at a large scale and can read, process, and write more than 150 GB of data in less than 4 minutes.

Google Bidding Automation

Fanatics is the largest online licensed retailer of fan gears for all major sports leagues merchandizing millions of products. Our team is responsible for bidding on those products to improve the visibility on google ads.

To achieve this, I have worked on designing and developing an application responsible for creating ads, updating bids, and deleting ads in Google ads. This application is developed in Spark Scala for data processing and Java for making api calls. To further improve the performance and enable dynamic task generating we used an orchestration tool: Netflix Conductor

Hot Market Google Performance tool

Worked on creating a Full-stack application that is critical to pull the google ads performance reports based on defined settings during the hot market. The frontend of the tool is developed in React, the backend application is developed in Python Flask and MySQL for database.