Portfolio management
I built the credit portfolio manager from scratch, covering the interface, backend, data and infrastructure. The stack included React, Elixir, Python, PostgreSQL and AWS.
← Back to projects
Data and credit · Professional experience
Bom.Capital: from a credit application to a response in under 2 hours
At Bom.Capital, a digital microcredit fintech, I first worked in full-stack development and later as a data scientist. As the company’s first data professional, I built the ML solution for credit analysis and approval.
- My role
- Full-stack development and data science
- The project
- Credit portfolio and application analysis
- The evidence
- Self-reported · no external publication
Time to a response was part of the problem
In my professional account, the process from application to customer response took an average of 48 hours. I worked on the operational software and later on the Machine Learning solution supporting credit analysis.
These are complementary efforts: a portfolio management system and a data layer supporting analysis. The reported improvement concerns process time, not portfolio profitability or credit quality.
My contribution
Two connected efforts, from application to model
The ML solution
As the first data professional at the company, I built the solution for credit analysis and approval, connecting data work to the business process.
Working across teams
I contributed to the early credit scoring model and communication with external teams and companies, connecting development with operational needs.
The process covered by the comparison
Application received
The starting point of the interval in my account.
Credit analysis
The stage supported by the Machine Learning solution.
Customer response
The end of the interval used for comparison.
An explanatory process diagram, not a reproduction of company screens or decision rules.
The outcome
From an average of 48 hours to under 2 hours
The work connected software and Machine Learning with credit analysis. In the process I worked on, customer response time fell from an average of 48 hours to under 2 hours.
Hours per process. Scale from 0 to 48 h.
Time from application to response
- Before · average
- 48 h
- After · reported
- < 2 h
The dashed bar marks the 2 h bound. The reported value is below it; no exact duration is available.
Source: Henrique Pett’s account, with no external publication. “Under 2 hours” is an upper bound; the bar does not represent an exact duration.
What I bring from this experience
A data solution must become part of the workflow people actually use. Working on both the portfolio system and credit analysis gave me experience connecting interfaces, rules, data and Machine Learning to a specific operational need.
Sources and context
Source: my professional history at Bom.Capital. Chart created from the reported values, not an independent audit. Experience predating the studio.
Case study prepared October 14, 2026.
Does your team wait too long for an analysis or approval?
We can map the workflow, find where work accumulates and define a delivery we can verify.