Skip to main content

flagship field record

Data Infrastructure Modernization

2022-2026Data EngineeringClient: TalkingPoints

I worked at TalkingPoints from 2022–2026. Each project lists its own dates, so some begin later within that span.

70% cost reduction, near-real-time data freshness
Problem
A young data platform had to become reliable and affordable as usage grew.
My role
Joined the early data team and grew into a senior data role while leading the modernization.
Shipped
Maintainable ELT, dbt architecture, CI/CD, and visible cost controls.
Proof · 2026-07
70% Snowflake cost reduction with near-real-time data freshness.

Public systems

Built from scratch (-ish). Hevo → (a few other things 😬) → Fivetran, Medallion architecture with dbt, CI/CD on GitLab, 70% Snowflake cost reduction.

Materials & technology

SnowflakedbtFivetranGitLab CI/CDMongoDBAWS S3Python

Links

Frequently Asked Questions

What made the Fivetran migration worth the effort?

The old ETL was a maintenance burden that required deep tribal knowledge to debug. When it failed, we had to dig through custom code to understand why. With Fivetran, failures are visible, documented, and usually self-recovering. The time we saved on maintenance went into building things that matter.

How do you approach Snowflake cost optimization without breaking things?

Start with usage patterns, not query plans. Talk to stakeholders about when they actually need fresh data—often the answer is 'not as frequently as we're providing it.' Then work backward to the cheapest architecture that meets real needs.

What would you do differently if starting over?

I'd push harder on stakeholder conversations earlier. We spent time optimizing queries that turned out to be unnecessary once we understood actual usage patterns. The technical work is the easy part—the people work is what determines success.