Independent data platform advisory

Better data platforms for rough terrain.

The expensive data problems are rarely a missing tool. They are a platform that grew up under different assumptions, a bill nobody can model, or a metric that means three things depending on who you ask. I work on those directly with the people who own them: architecture, cost, sequencing, and whether the numbers are safe to quote. One senior advisor, remote across the US and Europe. Fifteen years inside Zillow, Kin, Pluralsight, and Kustomer.

Architecture · platform economics · roadmaps · metric trust · data residency · operating model

Book a 30-min fit call Or send a note

I reply within 48 hours

82%

ETL cost reduction at Zillow

~50%

compute cut while the platform grew

one metric definition, every surface

US + EU

residency regions shipped and operated

Signs you're on rough terrain

Where I usually get called.

01
The platform was designed for a tenth of today's volume, and nobody wants to be the one who finds out which part gives first.
02
The cloud bill is growing faster than the business. Finance asks why, and engineering has a long thread but no model.
03
Revenue, or churn, or active users means something different depending on which dashboard you open. Everyone is tired of arguing about it.
04
Every data incident turns into an ownership question before anyone opens a runbook.
05
There is a seven-figure migration or rebuild on the table, and one chance to sequence it correctly.

If two or more of these are familiar, it is worth a conversation.

Selected work

Results you can check.

Two problems I owned end to end. Numbers first. The same class of work I take on now.

Make every number safe to quote

Kustomer

1x: metric definition, every surface
18-30ms: parity overhead on the serving path
Certified: line between quote-safe and exploratory

The warehouse was governed. The surfaces people trusted were not. Same business question, different answers in-app, in BI, and in finance, because every tile re-implemented the metric in hand-written SQL. Governance stopped at the gold layer.

Closed the loop. A self-hosted MetricFlow serving layer over ClickHouse so app and BI query the same definitions under multi-tenant security. Then a nightly quality score and Certified badge that only lands when a metric has a ratified brief, a semantic definition, and proven provenance on the surface. Models advise on failures. They never assign the score.

Write-once metrics for customer-facing reporting and internal BI
Multi-tenant RLS with byte-identical parity checks
Self-serve BI with certified finance, product, and CS surfaces
"Safe to quote" enforced by the platform, not a wiki

Read the full case for Make every number safe to quote

Cut the bill without cutting the platform

Zillow

82%: ETL cost reduction
43%: faster queries
21%: less engineering overhead

Fragmented stacks and rising ETL cost were slowing product and ML. Ownership stayed unclear.

Unified data science, MLOps, and engineering on a lakehouse path, tightened governance as ownership decentralized, and put real-time clickstream on the product decision loop.

One plane for science, MLOps, and engineering
Clearer ownership without weaker controls
Clickstream feeding product decisions in real time
Cost outcomes leadership could take to finance

Read the full case for Cut the bill without cutting the platform

These are operating outcomes from inside those companies, not invented client logos. New advisory work stays off the website unless you ask to be named. Diligence is easy: LinkedIn, the quotes below, and ask for a live reference on the fit call.

Engagements

How this works commercially.

Three ways to buy the work. You work with me directly, start to finish.

Terrain Survey

2-3 weeks

Fixed fee · typically mid five figures

A clear-eyed read before you spend real money. What is actually broken, what is theater, and which decisions matter. I am not here to bless a vendor shortlist you already picked.

A written assessment, a prioritized risk list, and a readout your leadership can use without a decoder ring.

Before a migration, a rebuild, or a tooling bet that will still be on the books in three years.

Route Plan

4-8 weeks

Fixed fee · scoped to the decision

Target architecture, sequencing, cost, build-vs-buy, who owns what. Written so engineers can ship it and finance can fund it. Idealism optional. Arithmetic required.

A roadmap with milestones, risks, staffing shape, and numbers that survive a skeptical CFO.

When modernization has to get past the board, not just past the architecture review.

Traverse

Monthly retainer · hard cap

Fractional principal · ongoing

I stick around for the hard calls: design reviews, hiring bar, vendor pressure, the weekly 'should we really do this' moments. Same person in Slack and in the exec meeting.

Ongoing judgment from someone who has held the pager, without paying for a full-time platform exec you are not ready to hire.

When the platform is live and the decisions do not politely stop arriving.

Advisory. Not staff aug, not a body shop, not a reseller with a services arm. Fixed fee or a capped retainer. Priced for the size of the decision, not the length of the deck.

A good fit

The decision is expensive if you get it wrong.
You want one senior person, not a slide factory.
A CTO, VP, or platform lead will actually show up.

Not a fit

You need seats filled by Friday.
You want someone to rubber-stamp a choice already made.
The main criterion is the lowest hourly rate.

How we start

Thirty minutes on the calendar. If it fits, a scoped proposal. If it does not, I will say so. Either way, you hear from me within 48 hours.

Working with US and EU clients

Remote across the US and Europe, on overlapping US and EU hours. Engagements run in English.
I have designed, shipped, and operated multi-region platforms under real data-residency requirements, including EU and India.
Two-week notice on scheduling changes, and a written summary after every working session.

Terms

Mutual NDA before any non-public technical detail. Yours or mine.
Client work stays confidential. Nothing appears on this site without written permission.
Fixed fee or a capped monthly retainer, agreed in writing before work starts. No open-ended time and materials.
Advisory only. I do not provide staff augmentation or resell implementation.

Who you work with

Hi. I'm Alex.

Comfortable in a code review and a budget meeting. Prefer neither to be theater.

Alex Silva

Engineer · Operator · MBA

I've spent fifteen-plus years inside companies where the data plane is the product... Zillow, Kin, Pluralsight, Kustomer, a few before that. I like the unglamorous work: making the bill make sense, making the metric mean one thing, making a migration survive contact with production. I also have an MBA from UCF, which mostly means I can sit with a CEO or a CFO without turning into a different person. Data Outfit is me, on purpose. No bench. No partner logo wall. If the work needs a squad of contractors, I will tell you that on the fit call and wish you luck.

Inside

Kustomer · Kin · Zillow · Pluralsight · Rackspace

“If you cannot explain the tradeoff to the people funding it, the architecture is not done.”

MBA

University of Central Florida

15+

years building data platforms

10-50+

engineers led per organization

What colleagues say

People who have worked with me, in their own words. The rest are on LinkedIn.

“Throughout the process, Alex demonstrated vast technical acumen, visionary leadership, and a commitment to continuous learning, which ultimately resulted in one of the most diligent evaluations I've experienced in my entire career. From the start, it was clear that Alex is proactive when attacking challenges, finding solutions, and setting his teams up for success.”

Nick MilanoData + AI at Databricks · Alex was Nick's client at Kin

“I tapped Alex to be the primary architect for the technical underpinnings of a transformational data infrastructure project. Alex was never afraid to dive into incredibly big, complex and technically challenging projects... He was a strong evangelist, educating and communicating with stakeholders across the company.”

Michael KowalchikManaged Alex at Pluralsight

Talks and writing

Selected conference talks on the systems behind this work.

Designing a reactive real-time data platform
Reactive Summit · 2017
Designing a reactive data platform: challenges, patterns, and antipatterns
O'Reilly Software Architecture Conference · 2016
Designing a scalable real-time data platform using Akka, Spark Streaming, and Kafka
Strata + Hadoop World · 2016

Full background and recommendations on LinkedIn →

Contact

Tell me what is on fire.

Or almost on fire. Thirty minutes on the calendar, or a short note. Problem, constraint, timeline is plenty. I reply within 48 hours. If it is not a fit, I will say so... saves us both a longer meeting.

Book a 30-min fit call Or send a note