I build the systems
behind products that scale.

Demand-side platform

Problem Advertisers need to manage complex campaigns while the platform connects to many ad exchanges and continuously controls pacing, delivery, optimization, and creative operations.

Approach Designed and developed the Spring Boot backend for campaign, insertion order, line item, creative, audience, deal, and conversion workflows, integrated with a distributed real-time bidding system handling 100K+ requests per second. The RTB path follows the OpenRTB protocol with low-latency responses, while scheduled services manage pacing, budget allocation, delivery tracking, KPI optimization, and notifications.

Centralized IAM platform

Problem Products needed one consistent way to sign users in, manage tenants, and enforce permissions across services.

Approach Built a Spring Boot identity service with Cognito-backed sign-in, multi-tenant user and company management, role/ACL-based authorization, signed platform JWTs, audit support, and event-driven billing lifecycle integrations.

Shared backend building blocks

Problem Every product team was solving the same cross-cutting concerns independently, creating duplication and inconsistent behavior.

Approach Created reusable versioned Java modules for authentication, caching, MongoDB access, email, shared DTOs, and utilities so new services could start with dependable platform building blocks.

Massive-scale identity resolution

Problem A legacy HTTP/MongoDB platform struggled to scale as identity data grew, requiring recurring operational intervention and periodic reduction of historical data. Identity data arriving from multiple sources used inconsistent identifiers, making it difficult to maintain a stable and scalable identity layer as volume grew.

Architecture Re-architected the platform as a Spark-native data system using Apache Iceberg and EMR-on-EKS, orchestrated through Airflow. The system performs identity resolution against a 65B+ historical mapping dataset while processing 10B+ incoming device records daily, resolving one stable internal ID per device and writing refined Iceberg tables.

Key engineering decisions Optimized data layout and joins using Iceberg Storage Partition Joins with zero OOMs or disk spills; implemented controlled, deterministic and idempotent ID-resolution processing to prevent duplicate assignments; tracked first-seen/last-seen and opt-out/privacy state incrementally; and automated Iceberg metadata cleanup and compaction to control small-file fragmentation and maintain stable performance.

Campaign Measurement

Problem Customers need to understand whether advertising activity translates into physical visits and measurable business outcomes.

Approach Built backend capabilities for measurement campaigns, report scheduling, insight delivery, and Spark orchestration that turns location data into footfall and attribution results.

AI-powered semantic search

Problem Users needed a more natural way to discover relevant content across large collections of blogs, audiences, and locations than keyword-only search could provide.

Approach Built a reusable semantic-search capability indexing 100K+ items using embeddings, vector retrieval, AWS Bedrock, and LLM-assisted query understanding.

Voice command recognition engine

Problem Final inspection teams needed a faster, more natural way to interact with the inspection workflow and reduce command latency.

Approach Prepared voice data with Python libraries, trained a machine-learning model to identify commands, exposed it through Python Flask REST services, and supported end-user training and go-live at a Nissan manufacturing plant in Oppama, Japan.

Vehicle logistics visibility

Problem Logistics teams need visibility into a vehicle’s journey from manufacturing plant to dealer across Renault and Nissan operations.

Approach Developed REST APIs and worked across UI and backend layers to support vehicle logistics workflows and improve information flow for users.