DataBackfill

Conversational AI Platform

Architected a multi-agent AI platform with tool-calling and long-term memory, improving first-contact resolution by 42% for over 1.8 million monthly conversations.

Developed a hybrid RAG pipeline using Vertex AI and BigQuery, successfully reducing hallucination rates by 58% through continuous automated evaluation.

Engineered high-throughput data pipelines with Apache Beam that processed 40k+ events/sec, enabling cost-aware model routing and reducing overall LLM costs by 31%.

Maintained 99.9% availability while implementing robust observability and production guardrails to ensure system safety and reliability under rapid iteration.

42%

RESOLUTION LIFT

58%

HALUCINATION REDUCTION

31%

COST SAVINGS

Go

Vertex AI

BigQuery

Apache Beam

Kubernetes

RAG

System Design

Google

Search & Workspace AI

Optimized core ranking for Google Search Query Suggestions, improving relevance and CTR for 150M+ daily users while strictly adhering to low latency budgets.

Delivered Gemini-powered Smart Compose features for Gmail, reducing average user composition time by 26% and increasing feature adoption significantly.

31%

RELEVANCE LIFT

26%

TIME SAVED

C++

Java

TypeScript

React

gRPC