Intent Accuracy

93%

Latency reduction

3.9m to 1.4m

Infrastructure

99.9% uptime

Data processing

Reliable systems

AI & LLM Integration

RAG · Agents · Orchestration

Data Pipelines

Beam · BigQuery · Pub/Sub

Conversational AI

Latency · Accuracy · Cost

System Design

Distributed · Scalable · Cloud

GCP

GKE · Cloud Run

Event Streaming

Kafka · Pub/Sub

Caching & Storage

Redis · PostgreSQL · BigQuery

Automation

CI/CD · Terraform

Led the architecture for a multi-agent AI platform serving over 1.8M monthly conversations, implementing model routing and tool-calling to improve first-contact resolution by 42%.

Designed a hybrid RAG engine using Vertex AI Vector Search and sparse retrieval. Achieved a 17% lift in conversion through continuous offline evaluation and online feedback loops.

Built high-throughput data pipelines using Apache Beam and Pub/Sub, processing 40k+ events per second while lowering average LLM costs by 31% through optimized model routing.

Managed core ranking components for Google Search, improving suggestion relevance by 31% and CTR by 19% for 150M+ daily active users while maintaining strict latency budgets.

Engineered personalized feed ranking services that reduced latency from 115ms to 47ms (p95), directly contributing to an 11% increase in session duration through iterative experimentation.

Implemented streaming generation for Gemini-powered features, reducing average composition time by 26% and ensuring high availability across global web and mobile clients.