Featured Case Studies
Four projects written up in full: the problem, the constraints, what I did and the result.
20+ enterprise clients migrated, zero data loss
Compliance records moved from Convercent to OneTrust with config-driven mapping, idempotent APIs and phased validation.
Read the case studyLLM chatbot in production in 6 weeks
LangChain, HuggingFace and FastAPI on AWS Kubernetes, answering questions over security data for 12M+ companies.
Read the case studyAzure to AWS with 35% less operational overhead
A legacy bioanalytical platform moved between clouds behind new REST APIs, without interrupting the labs.
Read the case studyCredit scoring engine used by thousands of companies
Led 6 engineers to deliver a PySpark and neural network decisioning engine for Brazil's largest credit bureau.
Read the case studyBackend and Integration
Media Migration Across Microservices
Challenge
Thousands of media records a day had to move between microservices and into Firebase for one of Brazil's largest media companies, while the products reading that data stayed online.
What I did
Built a migration pipeline on Django REST Framework that wrote to Firebase in batches. Every write was idempotent, so partial failures could be retried. The old read path stayed live until the new one was verified.
Result
Thousands of records moved daily with zero downtime for the products depending on the data.
QA Automation for Payment Transactions
Challenge
New financial transaction field types had to ship on one of the world's highest-traffic payment platforms without breaking the experience of millions of daily transfers.
What I did
Built Kubernetes QA automation and Python test suites for the new field types, including validation decorators that reject malformed input.
Result
The new transaction fields were covered by automated tests running in the platform's CI/CD pipelines.
WebSocket Load Testing for a Banking Platform
Challenge
Caixa needed proof that a new real-time banking platform could handle extreme concurrent load without degrading or dropping connections.
What I did
Engineered a distributed load-testing framework on AWS that simulated 30,000 concurrent operators and 100,000 virtual clients over WebSockets, with realistic behavior, variable connect and disconnect rates, and detailed metrics.
Result
The tests found critical performance issues before production. The suite became a standard part of Caixa's release process.
Data
Data Ingestion for Security Ratings
Challenge
Security data came from many sources with inconsistent formats and unreliable delivery, yet the ratings built on it had to stay accurate.
What I did
Built ETL pipelines that standardized the data, with automated validation, anomaly detection, reconciliation and backfills for source outages.
Result
The platform could rate more third-party vendors per customer while keeping the data accurate.
Wildfire Detection from Camera Images
Challenge
Images from remote cameras had to be analyzed to catch early-stage wildfires, with corrupted images and changing light producing false alarms.
What I did
Engineered fire detection APIs with TensorFlow, Keras and OpenCV, plus a pipeline that extracted visual features and validated images before detection.
Result
Detection latency dropped from 5 minutes to 30 seconds.
License Plate Recognition for Smart Cities
Challenge
A smart city platform had to recognize vehicles in real time across hundreds of city cameras, with high accuracy.
What I did
Built neural network pipelines with OpenCV, Keras and TensorFlow for license plate recognition, with Flask APIs and indexed storage for fast vehicle history queries. I also maintained the legacy Java Spring system and ran OWASP penetration tests.
Result
Real-time license plate recognition across hundreds of city cameras.
Real-Time Network Monitoring for a Bank
Challenge
A bank network with thousands of hosts needed real-time monitoring, with strict limits on alert response time.
What I did
Built a multi-threaded Django and SNMP monitoring system that collected device metrics continuously, stored history as time series and drew the network topology live.
Result
Real-time visualization of thousands of hosts on the bank network.
Cloud and Platform
Microservices for a Security Ratings Platform
Challenge
Growing data collection and API traffic were hitting the limits of a monolithic architecture, slowing deployments and scaling.
What I did
Split services along business domains, built resilient ingestion across data sources and added API versioning so clients could move over smoothly.
Result
Sole backend owner across three Django systems, shipping features weekly.
Kubernetes Deployment Tools for Client Environments
Challenge
The migration platform had to run in very different client environments, including air-gapped networks with strict security policies.
What I did
Designed containerized client tools on Kubernetes with encryption in transit and at rest, infrastructure-as-code templates and built-in logging and monitoring.
Result
Deployment time went from days to hours, including for the clients with the strictest security requirements.
Legacy PHP to AWS Serverless
Challenge
A legacy PHP system had to move to AWS, get faster and cost less to run.
What I did
Rebuilt it on AWS Lambda, S3 and API Gateway, tuned the functions for cold starts, added tiered caching to cut database load and built the CI/CD pipelines.
Result
The system ran serverless on AWS with CI/CD in place.
CI/CD for Data Processing Teams
Challenge
Data engineers spent too much time on deployments, and differences between environments caused bugs that were hard to trace.
What I did
Built CI/CD pipelines with GitHub Actions and Jenkins that test, validate and deploy data code, with Terraform for consistent environments and parameterized deployment templates.
Result
Deployments became repeatable across environments.
Have a Problem Like One of These?
Thirty minutes, no pitch. Tell me what is not working and I will tell you how I would approach it.