I'm an AI Engineer specialising in production-grade machine learning systems, LLM fine-tuning, and RAG pipelines. I build intelligent backends that turn unstructured data into measurable business outcomes.
I'm an AI Engineer, with 10+ years of experience designing and building intelligent systems — from LLM-powered chatbots and RAG pipelines to agentic automation tools and production web platforms. I've led the full stack from model experimentation to cloud deployment on AWS Bedrock and Azure Foundry.
I founded Estudar TI, an AI training and consulting company in Brazil, where I led a cross-functional team to ship 200+ projects and AI training programmes that reached over 7,000 students.
I'm currently pursuing a Master of Science in Computer Science at the University of the Potomac while staying hands-on with the latest developments in LLMs, agentic systems, and GenAI. I'm based in Silver Spring, MD and open to new opportunities in the US.
Technologies I work with:
▸Python
▸LLMs & GenAI
▸RAG & Agentic Systems
▸LangChain / HuggingFace
▸PyTorch / TensorFlow
▸AWS / AWS Bedrock
▸Azure AI Foundry
▸Google Cloud / Vertex AI
▸JavaScript / REST APIs
▸Next.js / React
▸SQL / Postgres, MySQL
▸Linux
02.
Where I've Worked
AI Engineer @ Estudar TI
May 2015 — Oct 2025
▸Designed and implemented AI agents, LLM-powered chatbots, RAG pipelines, and automation tools using LangChain, HuggingFace, PyTorch, and TensorFlow.
▸Deployed AI-powered applications on Amazon AWS Cloud, AWS Bedrock, and Azure AI Foundry, leveraging managed services to ensure scalability, reliability, and secure integration.
▸Built and maintained production-grade web platforms using Python, PHP, JavaScript, and MySQL, ensuring performance, reliability, and security across multiple high-traffic sites.
▸Maintained and enhanced a CRM and chatbot platform built with React, Node.js, and Redis, focusing on performance, scalability, and robust error handling in production.
▸Led a cross-functional team of 4 (design, marketing, sales, and engineering) to ship digital products, online courses, and AI trainings reaching more than 7,000 students.
PythonLangChainHuggingFacePyTorchAWS BedrockAzure AI FoundryReactNode.jsMySQL
03.
Featured Project
Enterprise SaaS · AI Platform
Upskill
An enterprise-grade AI learning platform that integrates directly into team workflows via Slack and Google Workspace — autonomously detecting skill gaps and generating personalized learning interventions using a dual-model pipeline (Llama 4 + GPT-5) on Microsoft Azure Foundry.
A modular fine-tuning framework for adapting Llama 3 and Mistral to domain-specific tasks using QLoRA and PEFT. Achieves 75% VRAM reduction via 4-bit NF4 quantisation, enabling full training runs on a single consumer GPU. Ships with W&B experiment tracking and a benchmarking suite covering perplexity, ROUGE, and LLM-as-judge evaluation.
An agentic, production-grade RAG system with hybrid retrieval (ChromaDB vector search + BM25) and cross-encoder re-ranking to cut hallucination rates on enterprise documentation. A LangGraph agent routes each query — answering from the internal knowledge base or falling back to GPT-4 / Claude 3.5 as needed.
A production-ready MLOps template covering the full model lifecycle — from automated training pipelines and model serialisation to a FastAPI inference layer, multi-stage Docker packaging, CI/CD via GitHub Actions, and real-time Prometheus metrics for latency and throughput observability.
A CNCF-hosted tool that tracks and surfaces the OpenTelemetry ecosystem — monitoring third-party integrations, SDKs, and plugins across hundreds of repositories via automated watchers.
PythonOpenTelemetryRefactoringuvCNCF
What I did
▸Architected watcher-common Python package with 3 shared base classes (VersionDetector, BaseInventoryManager, BaseRepositoryManager)
▸Eliminated ~280 lines of duplicated version detection and git/repository management logic
▸Wired the package as a uv workspace member so each watcher depends on it via { workspace = true }
▸PR merged by OpenTelemetry maintainers into this globally impactful CNCF project
Implemented LlamaCppLLM and CloudflareLLM Providers
About the project
An open-source Python framework for building LLM-powered applications with a unified interface across 23+ providers and 38+ tools, designed for easy extensibility.
PythonLLMAsyncIOCloudflareLlamaCPP
What I did
▸Built LlamaCppLLM for on-device GGUF model inference; used asyncio.to_thread() to keep streaming non-blocking
▸Built CloudflareLLM integrating Cloudflare Workers AI via the /ai/run/ REST endpoint with SSE streaming via httpx
▸Contributions part of the release that grew the platform to 23 providers and 38 tools
An open-source platform for automated student assessment using AI — providing flexible grading pipelines, rubric evaluation, and A/B testing of grading approaches for researchers and educators.
TypeScriptReactCSS HoudiniTailwindshadcn/ui
What I did
▸Built reusable AiInput and AiTextarea components with an animated glowing border effect using CSS Houdini (@property --ai-angle) and a rotating amber-to-cyan gradient
▸Components support ref forwarding, containerClassName, and full accessibility
▸Integrated AiTextarea into the rubrics dialog's 'Generate with AI' section, replacing a redundant inline gradient
An AI-powered learning platform that transforms chatbots into structured educational companions using cognitive science principles — covering 800+ subjects across 15 national education systems with 6 AI tutors and multi-agent classroom simulation.
TypeScriptNext.jsMarkdownAI Literacy
What I did
▸Authored a 342-line skill definition (SKILL.md) with a three-layer progression: AI User, AI-Enhanced Worker, and AI Builder
▸Implemented a diagnostic-first, Socratic teaching methodology with spaced repetition checkpoints across each layer
▸Registered the skill in tree-config.ts, wiring prerequisite (k12-mathematics) and downstream edges (data-analysis-stats, tech-career)
A collection of smaller experiments, competition notebooks, and side projects.
Year
Title
Description
Built with
Links
2024
YouTube Comment Fetcher
Agentic automation tool that fetches YouTube comments via the official Data API, then uses OpenAI to generate blog articles and video scripts from the content. Integrates with Heygen for AI video generation.
PythonOpenAIYouTube Data API
2024
WP Comment Responder
WordPress plugin that automatically replies to blog comments using GPT-4o-mini. Runs as a background cron job with configurable assistant behaviour, admin settings UI, and built-in update checker.
PHPWordPressOpenAI GPT-4o-mini
08. What's Next?
Get In Touch
I'm currently open to new opportunities — whether it's a full-time AI Engineering role, a consulting engagement, or an interesting collaboration. My inbox is always open.
