Staff Analytics Engineer · San Francisco

Gregory
Spathias

I build AI-powered data products and large-scale analytics systems.
Currently at LinkedIn — architecting intelligent sales infrastructure at the intersection of data engineering and LLM agents.

About

I'm a Staff Analytics Engineer with 7+ years building data infrastructure that drives revenue decisions. My work lives at the intersection of scalable pipelines, product analytics, and — increasingly — LLM-powered systems that put intelligence directly in the hands of sellers and operators.

At LinkedIn, I've designed and shipped data products across the full GTM stack: from Trino-based pipeline architecture and incremental loading patterns to a from-scratch AI agent platform for enterprise sales intelligence.

I care about building things that actually get used — which means obsessing over the last mile between a clean pipeline and a decision that changes behavior.

AI / LLM
LangChain LangGraph RAG architectures Streamlit OpenAI API
Data Engineering
Trino / Presto dbt Airflow Spark Snowflake
Languages
Python SQL Scala
Cloud & Infra
AWS GCP Kafka Kubernetes
Experience
LinkedIn
2019 — present
Staff Analytics Engineer, Marketing Solutions

Lead analytics engineering for LinkedIn's Marketing Solutions GTM organization, owning data infrastructure that informs revenue strategy across enterprise sales. Architected the NEON Scaled Customer Insights Agent — an LLM-powered platform that surfaces real-time account intelligence for enterprise sales reps. Designed and maintained Trino-based data pipelines at scale, including incremental loading patterns and event classification systems serving the full commercial sales org.

Trino dbt LangChain Streamlit Python Airflow GTM Analytics AI Agents
Projects
01
NEON Scaled Customer Insights Agent

An LLM-powered sales intelligence platform built for LinkedIn's enterprise sales org. Architected a five-step agent interaction flow (Query → Discovery → Insights → Analyze → Export) with account memory persistence, enabling sellers to surface real-time account intelligence without analyst bottlenecks.

LangChain · Streamlit · Python · LLM · RAG
02
GTM Event Attribution Pipeline

End-to-end classification pipeline for event-related ads at LinkedIn Marketing Solutions. Designed an extensible schema and LEFT ANTI JOIN incremental loading pattern in Trino, powering Q3 Events Crediting with a goal of full pipeline automation and revenue attribution accuracy.

Trino · SQL · Python · dbt · Airflow
03
LinkedIn Advertising AI Agent

A structured LLM system prompt and agent architecture enabling natural-language analysis of LinkedIn campaign performance data. Dynamically adapts to any campaign-related question, surfacing under-leveraged content topics and top-performing creatives by dwell time, CTR, and conversion metrics.

LangChain · OpenAI · Python · Prompt Engineering
Writing
Building production LLM agents at enterprise scale — lessons from NEON coming soon
Why LEFT ANTI JOIN is the right incremental loading pattern for most pipelines coming soon
Designing success metrics for AI-powered internal tools coming soon

Follow on LinkedIn for updates when articles publish.

Let's build something worth building.

I'm always interested in conversations about data products, AI infrastructure, and GTM analytics at scale. Whether you're working on something ambitious or just want to compare notes — reach out.