Gabriel Maestre Costa
Backend and full-stack developer
Most of what I do is for financial operations. I build the systems, keep them running, and keep changing them after they are in production.
The work is payroll-deducted loans, contract management, collection, protest, and credit recovery. Day to day that means business rules, data processing, and integrations between systems and services, including banks and platforms used by public agencies, financial institutions, and companies.
I also work across the stack, joining the frontend, the backend, and outside services. That includes async flows, recovering state, screens for processing data, and the links between applications and automations.
B.Sc. in Computer Science at UTFPR. During the degree I researched artificial intelligence, reinforcement learning, and multi-agent systems.
Ponta Grossa, Paraná, Brazil
app/api/profile/route.ts
Technologies
notes/technologies.txt
- Backend
- PHP / Laravel / Java / Spring Boot / Node.js / REST / SOAP / JPA/Hibernate
- Frontend
- Angular / React / Next.js / TypeScript / JavaScript / HTML / CSS
- Data
- PostgreSQL / MySQL / SQL / Flyway / QueryDSL
- Delivery
- Git / GitLab / GitFlow / Docker / CI/CD / Composer / Maven / SonarQube
- Practices
- Automated tests / Logging / Error handling / Authentication / Authorization / Auditing / Legacy systems
- AI
- Python / Reinforcement learning / Multi-agent systems / BDI / TensorFlow / Keras-RL / CARLA
Experience
Fácil Tecnologia
Two production platforms for financial operations. ConsigFACIL is a multi-tenant system for payroll-deducted loans: each client runs its own agreements, margins, contracts, and payroll integrations. ProFACIL handles credit recovery through protest and tax-debt certificates. Public and private creditors use it to bring debtors, titles, and notary offices into one operation.
AquiCob Soluções de Cobrança
A debt CRM for collection. The desk holds each debtor's balance, lets the team simulate and close agreements, issues invoices, and settles what was paid, with bank APIs on the financial steps. I also coordinated the technology team's backlog: what to build and in what order, including WhatsApp bots and outbound messaging for collection.
Moonrock Soluções Tecnológicas (contract)
I built the APIs and screens that connect business systems to each other. Backend services hold the domain rules, operators get the interfaces, and background jobs take the work that should not block the request.
Projects
Fig. 1 Scientific Initiation, Intelligent Agents
A DQN agent learned to drive in a simulated city. I set the control problem up in CARLA, ran the training loop in Python with TensorFlow and Keras-RL, and checked the agent in experiments as a deep network.
Fig. 2 Scientific Initiation, Traffic Simulation and AI
I compared CARLA, SUMO, and Traffic3D before later experiments settled on one of them. The choice used realism, behavioral fidelity, computational cost, and how directly each simulator connects to a reinforcement-learning stack.
Fig. 3 MASPY / Thesis
I compared a multi-agent learner with a monolithic one on the same intersection. In MASPY, traffic-light agents learn with Q-learning inside a BDI environment model. The baseline is tabular SARSA on an equivalent environment, and they share the transition, the reward, and the exploration.
github.com/Dev-Maestre/Smart-Queue-Agents-W-Maspy
Education
B.Sc. Computer Science, UTFPR, 2021-2026