Clínica Aurora
AI agent
A front desk request becomes a verifiable change in the clinic’s schedule. The agent finds the right appointment, asks for confirmation, and says so plainly when a rule or the clinic system blocks the change.
Salvador, Brazil (remote)
Backend and AI engineer. I build agents that write into your real systems, with the rules enforced in code.
Backend in Python and Node, integrations between systems, and AI where it solves the problem. From prototype to production.
Available for new contracts. UTC-3, with working-hour overlap with US Eastern.
About
I’m Pedro, a backend and AI engineer based in Salvador, Brazil, working remotely. I build applications, integrations and automations that connect a business’s systems and reduce manual work. My focus is backend development: APIs, databases, payments, background processing and platforms serving multiple businesses with separate data and permissions.
In AI, I build agents that retrieve information and take action in real systems: updating records, scheduling services, organizing requests and supporting operations through WhatsApp, Telegram or voice. I also develop document search and question-answering systems with sources attached to their answers, along with applications for transcription, translation and real-time audio.
I focus on how these systems behave during routine work. Business rules live in code, important actions can require human review, and decisions are recorded for inspection. My work covers spending limits, failure handling and clear paths for human follow-up.
I deliver everything from architecture and databases to the interface and the application running in production. My main stack is Python with FastAPI and TypeScript with Node.js, React and Next.js, supported by PostgreSQL, Redis, Docker and continuous integration. I use Claude Code and Codex during development, with review and testing to validate what I deliver.
Today I run Gomio, my software business, where I develop custom systems, automations and my own products. I’m also CTO of Peblink in South America, responsible for the technical direction of a platform currently in planning. I work on projects in Portuguese and English.
Selected work
Sofia
Voice agent
Gomio’s voice agent for prospective clients. Sofia explains Gomio’s services, learns what the caller needs, captures their contact details and sends the request to the team. Her decision console shows recorded events and the rules applied.
Chancela
SaaS product
Shows a local business whether AI assistants recommend it when customers ask, who they name instead, and which sources they cite.
Phone-menu mapping and live navigation
Client work, under NDAIn development
A consumer telephony app, not launched yet, that maps a company’s phone menu and guides a live call through it before handing the call over at the right department.
More work
Diária (opens in a new tab)
SaaS product
A reservations product for small inns with Pix payments and separate access roles for each property.
Live (in Portuguese): Diária (opens in a new tab)Code: Diária (opens in a new tab)
agent-ops
Python library
A Python library extracted from a document Q&A platform I built: a spend cap, a decision trail and a queue that survives a redeploy.
Gomio operation demos
Business automation
A real estate agency you operate from both sides: write as a tenant, then approve the reply in the owner’s panel. The same engine runs an accounting firm and a law firm.
Case: Gomio operation demosLive (in Portuguese): Gomio operation demos (opens in a new tab)
AI Web Scraping Agent
Data extraction
A URL and a sentence become validated JSON.
Case: AI Web Scraping AgentLive: AI Web Scraping Agent (opens in a new tab)Code: AI Web Scraping Agent (opens in a new tab)
Experience
Aug 2026 to now
Backend & AI Engineer (founder), Gomio
My software house for AI automations and conversational agents. Chancela measures business visibility in AI answers, Sofia handles voice inquiries about Gomio, and the operation demos illustrate business workflows.
gomio.com.brAug 2025 to now
CTO, South America (contract), Peblink
Pre-launch: the platform is still in planning, and there is no public product yet. Peblink runs workforce incubators and acceleration programs that connect students with real projects from companies and startups. I’m one of the partners; as CTO I’m responsible for the platform’s technical direction, and I lead the operation in South America.
Apr 2025 to Aug 2026, with backend work continuing into October
Backend & AI Engineer (contract), Consumer telephony startup (under NDA)
Backend for a consumer telephony app that maps a company's phone menu and navigates a live call to the requested department. The FastAPI services combine menu exploration, live call navigation, streaming transcription, telephony session control and integration with the app API. Interrupted or incomplete mapping runs cannot replace a better route; live navigation confirms sent key presses, rejects options the menu did not offer, and hands the call back when it reaches a person or cannot continue. Sensitive digits are redacted before they reach logs.
Read the caseOct 2024 to Jul 2025
Backend Developer (contract), Ehchad
Backend of a live speech-to-speech translator for conversations in English, Spanish and French. I moved the server off a desktop app onto a headless server, built the browser client that streams the microphone over Socket.IO, and rewrote the server on RealtimeSTT with Silero VAD and Faster-Whisper while keeping the existing client compatible.
Education
Software Engineering at UCSAL, in progress. Full-stack certificate from Springboard, 800+ hours.
How I deliver
How a contract starts and runs. Every step below is something I do today.
01
Diagnostic call
A free call to understand the process, the systems it touches and what a good result looks like.
02
Written scope
Scope, price and timeline in writing before any work starts.
03
Short pilot
A small first delivery on your real case, so you can judge the work before the full project.
04
Delivery
The full project, tested against realistic cases with the real model before it goes live.
05
Support
A support period after delivery for fixes and adjustments.
Engineering practices
- Business rules live in code.
- Deadlines, opening hours and limits are checked by code, not by the model, and each decision is recorded with the rule behind it, so you can check it later.
- Evaluated with the real model before release.
- Gomio’s operation demos are evaluated with the configured model, and Sofia’s scripted scenarios must pass before deployment.
- AI spending has a ceiling.
- Each system has a spending limit in code. When it reaches the limit, it stops calling the paid model instead of running up the bill.
- Releases are built outside your server.
- GitHub Actions builds each version, while the server downloads and starts the finished image. This keeps compilation off the production machine; pulling and starting an update still use server resources.