Well, hi there.
I'm Osvaldo. I like building software and experimenting with Agentic workflows using Amazon Bedrock AgentCore, and hopefully you're here because you do too.
I'm a Software Engineer focused on the frontiers of AI. This page is my digital twin, designed to answer questions about my professional background and technical interests.
We're applying AI to fields where it can make a massive, positive impact. Feel free to Connect with me for more!
Key Highlights
- Deployed a multi-agent RAG chatbot (this site's digital twin) using AWS Bedrock AgentCore, with source code on GitHub.
- Built agentic workflows on AWS EC2 with n8n, integrating ECS, Lambda, and vector databases for scalable AI automation.
- 10+ years engineering at Microsoft, Intel, Oracle—contributed to Visual Studio, VS Code extensions, and secure embedded systems.
- Completed MIT's Applied Agentic AI course and AWS certifications in GenAI and MCP.
Projects
Digital Twin Chatbot
This AI-powered chatbot is a multi-agent system built with AWS Bedrock AgentCore and deployed on Bedrock AgentCore Runtime, which provides security, governance, compliance, and observability via CloudWatch. It integrates AgentCore Gateway for tools, AgentCore Memory, Bedrock Guardrails, AgentCore Evaluations, and Bedrock Prompt Management. The knowledge base uses S3 with semantic chunking to answer questions about my background, projects, and AI expertise. There's room for future enhancements like streaming, model parameter tuning, and long-term memory strategies. Try asking about my agentic workflows or Bedrock integrations! Source code: GitHub Repo.
Digital Twin
Sign In to Access Digital Twin
Self-host n8n server
on AWS EC2 with Elastic IP for my personal workflows and automation with whatsapp and telegram. Live instance available at http://52.54.19.92:5678/.
Open-Source Contributions
Visual Studio Project System, VS Code C# Dev Kit, Jupyter Notebooks. See my GitHub.
Multi-Agent Conversational AI Assistant
Inspired by the MACRS paper (Multi-Agent Conversational Recommender System), this n8n agentic workflow creates a customizable "digital twin" – a personal AI assistant powered by multi-agent reasoning with LLMs. It includes context engineering (fetches conversation history, client profiles, strategy guidance), multi-agent response generation (Chit-Chat, Info Gatherer, Profile generators), refinement/guardrails (safety reviews, polishing), RAG ingestion (personal docs into vector store), and an evaluation loop (logs sessions, runs LLM-as-a-Judge).
Setup involves databases like PostgreSQL/Redis/Quadrant, credentials for personal details, and file paths for RAG documents. Ideal for building conversational bots for lead qualification, job inquiries, or personalized chats.
View Inspiring PDF | View n8n Workflow JSON
LLM Data Analysis Agent with Secure Docker Sandbox (MCP)
This project implements a sophisticated Data Analysis Agent that uses a Large Language Model (LLM) to perform data processing, statistical analysis, and visualization within a secure, containerized Python sandbox.
The system leverages the Model Context Protocol (MCP) to bridge the LLM with a dedicated execution environment, ensuring that code execution is isolated and safe.
View on GitHub
Package LLM Models for Ollama
Guide and tools for downloading, packaging, and running custom large language models (LLMs) with Ollama. Includes step-by-step recipes, Jupyter notebook examples, and Modelfile templates for models like Qwen2.5.
View on GitHub
LLM Digital Twin: Fine-Tuning Workflow
A comprehensive pipeline for fine-tuning a Qwen2.5-8B-Instruct model into a personalized "Digital Twin." This project demonstrates advanced LLM optimization and data engineering techniques, including:
- Fine-Tuning: Optimized LoRA training using Unsloth for 2x faster performance and reduced VRAM footprint.
- Data Engineering: Automated Q&A generation and embedding-based contradiction detection using Sentence-Transformers.
- Deployment: Production-ready FastAPI server and Ollama integration for local inference.