Alejandro R.

In this project, I led the development of a Multi-Agent RAG (Retrieval-Augmented Generation) application specialized by departments, designing and configuring the entire architecture and networking in AWS from scratch. For the AI engine, I used FastAPI and LangChain, implementing a data retrieval system based on PostgreSQL with pgvector for embedding storage and relying on AWS Bedrock and OpenRouter for response generation. Additionally, I complemented the solution by creating workflow automations for various departments, using both n8n and custom code implementations when the logic required more granular control. The result was an ecosystem capable of orchestrating multiple agents and automatic processes to manage departmental information precisely and efficiently.

During this period, I continued evolving the ETL backend engine, while also taking on the end-to-end development of a Power BI-like data visualization tool. I used Next.js, Symfony, and PostgreSQL to build an interface capable of generating dynamic dashboards by connecting to various real-time data sources. On the infrastructure side, I was responsible for modernizing the software lifecycle by implementing CI/CD pipelines with GitHub Actions, Jenkins, and AWS. This included scalable deployments on AWS Lambda and Fargate, utilizing SQS queues for asynchronous processes, and advanced network configuration to ensure security and performance. All development was managed under SCRUM methodology, maintaining constant technical communication with the client to align functionalities with their business needs.

In this project, I worked on the core of an ETL software, responsible for the processing logic and data engine architecture. I used PHP 8 and Symfony to develop efficient ingestion and transformation pipelines, always prioritizing system performance with large data volumes. A fundamental part of my role was advanced PostgreSQL management, focusing on data modeling, complex query optimization, and the implementation of indexing strategies to reduce bottlenecks. Additionally, I integrated solutions for background task processing and message queues, ensuring the system could scale and process data asynchronously with complete reliability.