Engineering decisions with verifiable results, in CAR format: Context, Action, Result and Lesson.
Claude Code skills that take a PEDIDO.md to a running system, and a script, not the agent, approves every stage.
5/5 requests with 100% of acceptance criteria met, 4 of them unseen
A single MCP server with per-user permissions and auditing that became the front door to the company's agent tools.
560+ automated tests in the gateway
From a Streamlit prototype to an ADK agent with A2A on Agent Engine: every architecture change answered a real problem.
7 architectures iterated over 7 months
A multi-agent system with prompt-injection defenses validated the need. The final architecture differed, and I built what it relies on.
3+ specialist agents, each with its own RAG corpus
A WhatsApp bot for receivables advance: the AI runs the conversation and a pruned beam search picks the invoices.
An academic MVP: two specialists with separate RAGs and a consolidator with structured output that feeds a dashboard.
Ongoing research: do humanities-based creative operators beat high temperature on originality, without losing appropriateness?
Smaller, academic or non-AI projects.
Day-shift marketplace for hospitality and events, with two-sided QR check-in. Co-founder; Flutter app, KYC and payments.
Website and store for a DJ gear reseller: SDD, prioritized feature catalog and prototypes.
Website for a fictional digital-inclusion NGO (SDG 9), a web programming project.
Public repositoryFront-end web course assignments.
Public repositoryCode portfolio for the Data Structures course.
Public repository