AI AUGMENTED ENGINEER? 
AI engineer shipping LLM-powered features, agent pipelines, and the interfaces that put them to work.
AI COMMODITIZED TYPING.
IT DID NOT COMMODITIZE ENGINEERING.
The engineer owns the thinking.
Understanding the problem. Exploring the solution space. Choosing the architecture. Weighing performance and cost. Designing the guardrails that catch regressions before users do. None of that gets delegated. AI accelerates the keystrokes, not the judgment.
Hidden decision debt is real.
Skipped judgment surfaces later as spec drift, performance regressions, and untraceable bugs. I keep decisions explicit so the debt stays visible.
Coherent output ≠ correct output.
LLMs are probabilistic compilers — they optimize for coherence, not correctness. Verification is structural: tests, types, guardrails. Not vibe-based.
Workflow is method, not ritual.
The tools change. The agents change. The pipelines get renamed. What stays constant is who's accountable for the result.
The signature I sign on the work is the judgment behind it.
That's the part worth paying for.
TRIAGE
Intake & routing
- AUTOMATED
Pull from Jira
Dev-ready tickets flow into the queue.
- AGENT
Read session-handoff.md
Restore full context per active worktree.
- AGENT
Classify complexity
Route each ticket — simple vs. intermediate.
- SYSTEM
Prioritized queue
Ordered & ready — fork by complexity.
EXECUTION
Two paths, by complexity
- HUMAN
Scope the change
Quick alignment on intent and approach.
- AGENT
Implement → PR
Agent codes the change and opens a pull request.
- HUMAN
Template + system design
You architect the approach before any code.
- AGENT
Taskify → isolated worktree
Break into tasks; spin up a clean worktree.
- AGENT
Sub-agents per phase
Phase-scoped agents write code against AC tests.
- AGENT
Agent review → manual validation
Self-review pass, then you sanity-check.
- HUMAN
Human review
Final read before anything can merge.
- AUTOMATED GATE
Guard rails CI
Any PR that regresses a metric gets automatic feedback — before it reaches human review.
- MILESTONE
Merge to main
Gated, reviewed, green.
CLOSING
Merge, teardown & capture
- AUTOMATED
Done
Remove handoff + tear down the worktree.
- AGENT
Still in progress
Write session-handoff.md for the next run.
- SESSION
Vault Session
Capture the key decisions.
- SESSION
Wrap-Up Session
Process improvements.
- SESSION
Learnings
Technical takeaways.
↻ next run
Products I've
built & shipped
Each project documents not just what was built — but why. The tradeoffs, constraints, and decisions behind it.
REVLA
Conversation intelligence platform for B2C teams
Fullstack SaaS dashboard that analyzes client–attendant conversations to anticipate risks and opportunities with AI. Shipped the CSV import that unlocked the company's first signed contract.
VEGAJUS
Legal-tech SaaS for Brazilian solo lawyers
Co-founded CRM + WhatsApp automation + Meta Ads platform. I lead frontend and product; co-founder owns backend. Vibe-coded validation prototype, then rebuilt v1 deliberately on modern stack.
COLLAMAP
AI-powered research suggestion engine
Main frontend engineer for 3 years on a multi-tenant research platform. Diagnosed a retention problem through engagement-data analysis, then shipped a 3-stage AI pipeline at 66% cost reduction in 4 days.