Projects

Selected work — architecture, systems programming, and applied AI.

• uLLAMA: A Minimal C Runtime for Local LLM Inference

  A small C wrapper around llama.cpp that loads GGUF models and runs an interactive session with live tokens-per-second reporting — built to fit an 8 GB, CPU-only laptop and written under the 42 Norm.

• Typed Decisions vs. Prose: A Support-Ticket Triage Demo

  A live, side-by-side demo of two ways an AI model can triage support tickets — fast typed decisions versus slower generated prose — built against a published model contract and localized for Brazilian Portuguese.

• Self-Correcting AI Teaching Activity Generator

  A web app built in 48 hours for the IBM Bob 2.0 Hackathon that generates hands-on, teacher-editable classroom activities for any resource level — with a second AI pass that audits its own output against six quality checks before the teacher sees it.

• Digital Resilience: Failover-Resilient Serverless Architecture

  A high-availability AWS architecture with dual-origin failover, built and live-tested for the EDN Hackathon Future Builders (Case B: Digital Resilience).

• Gems & Landi: AWS Cloud Labs Documentation Series

  An ongoing, self-published documentation standard for hands-on AWS labs — networking, security, and serverless — written with more depth than the official docs.

• Systems Programming at 42 São Paulo

  Reimplementing the C standard library, building a line-by-line file reader with static memory, and co-developing a stack-sorting algorithm — the foundations most cloud engineers skip.