Featured Projects
My main projects
For contractual reasons, this project cannot be exposed beyond what is described here.
Fullstack Platform
Fullstack & Mobile
About the Project
I work as a fullstack developer maintaining and evolving a production platform with nearly 1000 connected devices.
The platform centralizes the management and monitoring of these devices, and the mobile apps (Android and iOS) communicate directly with the embedded devices via BLE.
The frontend is built with Next.js and the backend with NestJS following DDD, Clean Architecture and CQRS, with Docker and CI/CD through GitHub Actions, and the whole infrastructure runs on AWS (RDS, IAM, Lambda, EC2, DynamoDB, S3 and ECR).
For contractual reasons, images, videos or further details about the product cannot be disclosed.
The platform centralizes the management and monitoring of these devices, and the mobile apps (Android and iOS) communicate directly with the embedded devices via BLE.
The frontend is built with Next.js and the backend with NestJS following DDD, Clean Architecture and CQRS, with Docker and CI/CD through GitHub Actions, and the whole infrastructure runs on AWS (RDS, IAM, Lambda, EC2, DynamoDB, S3 and ECR).
For contractual reasons, images, videos or further details about the product cannot be disclosed.
⭐
For contractual reasons, this project cannot be exposed beyond what is described here.
ISP Automation Platform
Automations, Cronjobs & AI
About the Project
A fullstack customer service automation ecosystem I built and maintained for an internet service provider (ISP): several services running in Docker containers on the same VPS.
At its core is an LLM-powered WhatsApp bot integrated with Chatwoot covering sales and support — it identifies the customer, sends PIX and bank slips, grants trust-based unblocking, analyzes the router photo with computer vision, restarts the equipment through the OLT and opens typed technical tickets. Below it, microservices implement the same API contract for different ERPs, with a router that directs each call to the system where the customer actually exists.
Around that, eight scheduled jobs: post-service follow-ups on D+1, D+7, D+15 and D+30, discovery of conversations from the service orders closed each day, reply polling, 5 and 15 minute synchronizations, alerts and reports — plus a multi-tenant dashboard that analyzes conversations with AI and flags churn risk.
Everything is layered (controllers → services → repositories) with dependency injection, a ten-pillar CI quality gate and unit, integration and end-to-end tests. I also administered the VPS: Docker Compose per service, deployment through Jenkins and GitHub Actions, healthchecks, log rotation and a rollback runbook.
For contractual reasons, images, videos or further details about the product cannot be disclosed.
At its core is an LLM-powered WhatsApp bot integrated with Chatwoot covering sales and support — it identifies the customer, sends PIX and bank slips, grants trust-based unblocking, analyzes the router photo with computer vision, restarts the equipment through the OLT and opens typed technical tickets. Below it, microservices implement the same API contract for different ERPs, with a router that directs each call to the system where the customer actually exists.
Around that, eight scheduled jobs: post-service follow-ups on D+1, D+7, D+15 and D+30, discovery of conversations from the service orders closed each day, reply polling, 5 and 15 minute synchronizations, alerts and reports — plus a multi-tenant dashboard that analyzes conversations with AI and flags churn risk.
Everything is layered (controllers → services → repositories) with dependency injection, a ten-pillar CI quality gate and unit, integration and end-to-end tests. I also administered the VPS: Docker Compose per service, deployment through Jenkins and GitHub Actions, healthchecks, log rotation and a rollback runbook.
For contractual reasons, images, videos or further details about the product cannot be disclosed.
⭐
Judite
Frontend & Backend
About the Project
Judite is a web platform created to automate business customer service, integrating WhatsApp, Messenger, Instagram Direct and Webchat through Meta's official API. The system offers automatic responses, scheduled messages, a dashboard with metrics, contact management and complete configuration of AI agents, including personality, tools and operating parameters. The platform also supports dynamic context extracted from videos, websites, PDFs, Word documents or free text, in addition to integrations with external services such as Correios, Google Sheets, Freshdesk, Brasil API and Asaas. The agent also uses a vector database, allowing more contextualized, accurate responses aligned with the company's products and services. With modular architecture and DevOps practices, Judite is a scalable, efficient and easy to maintain solution.
⭐
RTSP Detector
Classification, Detection and Facial recognition
About the Project
The RTSP Detector is software that transforms any security camera into an alarm center via RTSP protocol; it uses a real-time classifier that identifies images as "Fire", "With person" or "Without person"; if the classifier determines that some frame contains a person, a second model, this time a detector, is called to confirm the person's presence; if the presence is confirmed, a third detector is activated to identify if that person is carrying a firearm or a knife; the software also features facial recognition, allowing authorized people's faces to be registered so the system can tell residents and employees apart from unknown people; in addition, the software has a fire detection and classification system; if any event is detected, a message is automatically sent to the user's WhatsApp with the exact image of the frame where the event occurred; all models were trained by me, and all the interface and backend were developed with HTML, CSS, JavaScript, Electron and Python.
⭐
