Senior fintech product leader building AI systems that turn noisy inputs into reviewable decisions.
I lead product teams in fintech and build hands-on AI/product systems for signal curation, evaluation, local-first automation, and human-in-the-loop review.
My bias: AI becomes useful when it is connected to a real workflow, grounded in source evidence, and constrained by clear human approval boundaries.
| System | What it proves | Status |
|---|---|---|
AI Signal Desk |
A live AI signal product that turns noisy AI news, tools, repos, and concepts into practical calls: learn, try, watch, or ignore. | Live at aisignaldesk.ai |
skval |
Deterministic + LLM-assisted evaluation for Claude Code skills, with safety gates and ship/revise/reject scorecards. | CLI + docs site |
vyno |
Local-first AI digest pipeline with source curation, scoring, Telegram delivery, Obsidian archiving, and an operator console. | Personal automation |
shotback |
Human-in-the-loop visual QA workflow for screenshot capture, annotation, and LLM-ready product feedback. | Chrome workflow |
loopy |
Reusable agentic maintenance loops with deterministic detection, guardrails, and reviewable outputs. | Agent ops toolkit |
Most AI products fail in the gap between impressive output and trusted operation. I’m interested in the operating layer in between:
- Signal curation — reducing information overload into useful next actions.
- Evaluation and quality gates — making AI-assisted work testable instead of vibe-checked.
- Human approval loops — AI drafts, ranks, and proposes; people decide.
- Local-first automation — useful systems that remain inspectable, portable, and permission-aware.
- Product review artifacts — preserving source context, visual evidence, decisions, and tradeoffs.
These are useful context, but not the center of my current portfolio:
firehose— an earlier spec-driven workflow method for aligning AI coding agents with product intent.hermes-product-teams— a product-memory prototype exploring discovery notes, decision logs, PRD proposals, and weekly product briefs.sandy— a schema-first mobile UI prototyping sandbox for design-system and server-driven UI experiments.
I’m a Group Product Manager / Sr. Manager at Neon in São Paulo, with previous fintech and product leadership experience at Mercado Libre, Leve, PagSeguro PagBank, and ConectCar.
I use GitHub as a public workshop for practical AI/product systems: small enough to inspect, real enough to validate, and opinionated enough to show how I think.
- Start with the user workflow, not the model.
- Keep evidence attached to generated claims.
- Prefer reversible, inspectable systems over opaque automation.
- Use deterministic checks and fixtures where possible.
- Treat agents as collaborators that need context, constraints, and review.
If you’re building at the intersection of product leadership, fintech, and practical AI systems, I’m happy to compare notes.