A close that lives in Excel
Three days a month copying, pasting and checking cells. From scratch every time, and the error only surfaces after the number has already gone up the chain.
Data engineering · São Paulo
Reports that warn you when the data goes wrong.
I turn fragile, manual data into a number your board trusts without double-checking. Data engineering, executive dashboards and report automation, built by the architect who writes the code, with 19 years in financial-grade data.
A human reply · a diagnostic before any build · mutual NDA
Experience built at
BradescoSantanderSinqiaClaro
The path of your datareliable
Validated at every step, from the source system to the report.
1
Collect
Pulls from your ERP, database and spreadsheets. No manual copy-paste.
2
Validate
Checks the rules. Bad data is blocked here, so it never reaches the report.
quarantinethe error stays trapped here instead of becoming a report
3
Deliver
Becomes a dashboard and report the board opens without second-guessing.
The right number arrives ready. No spreadsheet that one person has to update by hand.
01The problem
The number reaches the board. But nobody can swear it's right.
When the data depends on manual steps and fragile processes, the error is silent. It shows up late, in front of the people who decide. Recognize any of these?
A dashboard nobody trusts
Every team walks into the meeting with a different number. When the board asks where it came from, nobody can answer with confidence.
A pipeline that breaks quietly
A load fails, a file layout changes, a column shifts, and nothing warns you. The report comes out clean and wrong.
A manual close, every month
The same spreadsheet ritual on repeat. The team is hostage to the routine, with little time left to read what the number is actually telling you.
Data scattered everywhere
ERP, database, spreadsheet, legacy system. Every decision starts with a hunt for the data before any real analysis.
Hostage to the master spreadsheet
One person understands the spreadsheet that holds the close together. If they take a week off, the process stalls with them.
02The solution
Reliable data isn't luck. It's architecture.
I rebuild the path of your data in clear stages, each one validated and recorded. You get a number that stands on its own and a team that understands how it was produced.
Diagnostic
Architecture
Ingestion
Validation
Transformation
Dashboard
Automation
Documentation
Support
01
Contract-first
The pipeline starts from a data contract, not from code. When the source changes, the contract catches it first and everything else reacts in a controlled way.
02
Separate layers + quarantine
Ingestion, validation and transformation in distinct stages. Anything outside what's expected goes to quarantine, not to the board's report.
03
Documentation as a deliverable
Runbook, data dictionary and recorded decisions are part of the scope. Your team can evolve the system without depending on me forever.
03Services
Six ways to make your numbers trustworthy.
Every engagement starts with a diagnostic. From there we scope the right work together, so you never buy blind.
04How I work
A diagnostic before any build.
A predictable process, from first contact to support. You know what comes at each step and why it's there.
- 1
Diagnostic
A 30-minute call and a read of your situation. Where it hurts, what already exists, what's at risk.
- 2
Mapping the sources
I map where the data comes from: ERP, databases, spreadsheets, APIs. Each source gets an expected contract.
- 3
Architecture design
I define the layers, the validation rules and the quarantine point before writing the first line.
- 4
Layered build
Ingestion, validation and transformation kept separate. Each step is testable and isolated from the rest.
- 5
Validation and quarantine
Data outside the contract is blocked and flagged. The number only reaches consumption after it passes the rules.
- 6
Documented delivery
Runbook, data dictionary and recorded decisions. The delivery includes what your team needs to run it.
- 7
Follow-up
A support and fine-tuning period, with 30 days of free bug fixes after delivery.
05Experience and authority
The person who sells is the one who delivers.
Tiago Pereira da Silva — Data Architect, 19 years in data.
I spent the last five years building the analytics layer of a bank in PostgreSQL, SQL Server, BigQuery and Power BI, with Random Forest and Bagging models in production. Before that, I served Santander in New York, Madrid and London through Sinqia, and worked at Bradesco and NET/Claro.
Pursuing an MBA in Data Science and Analytics at USP. Author of the open-source library InstaT (Python), with a test suite and CI. My experience is in financial services, where the wrong number carries a real, visible cost.
The difference is what breaks in production.
I once rebuilt a pipeline after a silent file-layout change shifted columns and corrupted a daily report with no one noticing. I split ingestion, validation and transformation and added quarantine. That kind of judgment, knowing where data actually fails, goes into every project.
19
years in data engineering
5+
years building a bank's analytics layer
3
countries served: New York, Madrid and London
100%
focus on mission-critical-number environments
Stack in use
PostgreSQL
SQL Server
BigQuery
Snowflake
Databricks
AWS
Power BI
dbt
Python
AirflowGEGreat Expectations
06What sets TPS apart
Why work with TPS.
The seller is the deliverer
You talk and work with the architect who writes the code. No sales layer, no handoff to a junior.
Documentation is part of the delivery
Runbook, dictionary and recorded decisions. Not an extra billed later. It comes with the work.
A diagnostic before any build
First I find where it breaks, then I build. You don't pay for a project defined in the dark.
Reliability first
The goal isn't the prettiest dashboard. It's the number the board trusts without checking it by hand.
Technical and business view
I understand the pipeline and I understand the close. The conversation is about the decision, not just the tech.
Straight talk
No empty jargon, no promise that won't hold. If something is a risk, I call it a risk.
Diagnostic
Want to know where your reports might be breaking in silence?
In a 30-minute call I show you the most likely risks in your data flow and what to fix first.
Book a data diagnostic
A human reply No commitment until the proposal Mutual NDA
07Frequently asked questions
Before you ask.
Didn't find your question? Send a message. I answer directly, no long form.
Yes. A lot of high-impact work is small in scope: automating a report that takes three days, or adding validation to a close. The diagnostic helps find the right first step, regardless of size.
Yes. In most cases I don't start from scratch — I start from your ERP, database, spreadsheets and current reports. The work is to organize and make reliable what you already have, without replacing everything at once.
It's one of the most common requests. I map what the spreadsheet does, rewrite the logic into testable code and automate generating the result. The spreadsheet stops being the critical system that depends on one person.
Both, usually together. An executive dashboard only matters if the number is reliable, and that depends on the ingestion, validation and transformation pipeline that feeds it.
Yes. Modeling, DAX and a real source connection. I also rescue Power BI dashboards that were deployed but no one uses, because no one trusts the data behind them.
Only what's necessary, and always under a mutual NDA before any detail. For the diagnostic, talking and seeing report examples is usually enough. Data access is agreed step by step, with the care of someone who worked in financial services.
Yes. Work is remote from São Paulo, with experience working with teams in New York, Madrid, London and, today, Germany. International payment via Wise (EUR/USD).
With a 30-minute conversation, no commitment. From there, I define a diagnostic with the map of the problem, the technical recommendation and the priority order. Pricing is presented in the proposal.
08Next step
Reliable data before the decisions that matter.
Tell me in two lines what bothers you most about your data flow today. I'll reply with an honest read and the next step.
- A human reply: you talk straight to the architect.
- A diagnostic before any build or charge.
- A mutual NDA before any sensitive detail.