I'm a Software Developer who ships end-to-end products, from database and API to frontend and production infrastructure. Backend in Python and Node is my core: APIs, auth, payments, real-time, background jobs, and multi-tenant architectures.
Most of what I build has AI at its core: multi-agent systems, RAG, voice pipelines (real-time STT and TTS, voice cloning), and vision workflows for documents and images. I wire multiple LLM providers together with failover, and run autonomous WhatsApp and Telegram agents in isolated multi-tenant infrastructure with persistent memory.
Frontend in React, Next.js, Tailwind, Framer Motion; infrastructure on Docker, Traefik, AWS, CI/CD.
I build daily with modern agent coding tools (Claude Code, Codex) and the Anthropic and OpenAI SDKs.
I run Gomio, a one-person software house in Salvador. If you have a process that should be running without a person in the loop, tell me which systems it plugs into and let's build it. See what I sell.
Time in Salvador, Brazil: --:--:--
Based in Salvador, Brazil, open to remote work and global collaboration.
Mail:contato@pgdev.com.br
AI-powered products that solve real business problems, from concept to deployed software.
AI assistants that take on your everyday business tasks, like answering customers, organizing information, and handling repetitive processes. They work on their own, around the clock, freeing your team to focus on what really matters.
Turn your company's documents, PDFs, and knowledge base into an AI-powered search engine your team can talk to. Hybrid search with semantic + keyword matching, reranking, self-correcting pipelines, and multi-tenant isolation.
End-to-end SaaS applications from authentication to payment processing to production deployment. I build the frontend, the API, the database, the billing, and the infrastructure, then put it in front of real users.
AI agents on WhatsApp and Telegram: support, booking, lead qualification and human handoff. Each client runs isolated with persistent memory, on the official WhatsApp Cloud API.
Speech-to-speech translation, voice-enabled document Q&A, real-time transcription platforms, and audio intelligence. Low-latency WebSocket streaming, VAD, and multi-provider TTS/STT integration built for production.
Production-grade REST APIs with auth, rate limiting, usage analytics and webhooks. Reliable Stripe and MercadoPago billing with metered usage and idempotent webhooks.
My Works
Open demo, no signup: you talk, and the row changes in the ERP live. Five business rules are decided by a deterministic engine instead of the model, every decision is logged with the rule code that produced it, and a button lets you take the ERP down to prove the agent escalates rather than inventing a confirmation.
Live demo
Three things every agent in production needs and nobody builds before getting burned: a spend cap that refuses when it cannot read itself, an audit trail that answers "why did it decide that" with a rule code instead of generated prose, and a durable queue that does not lose work across a redeploy.
Read the code
The same clinic schedule, published as an MCP server so another agent can plug into it. Every write takes two calls: the first returns a proposal and a token without touching anything, the second commits it. A model talked into skipping the confirmation still cannot.
Upload your PDFs, ask out loud, hear cited answers streamed back as natural speech. End-to-end voice loop: mic capture, gpt-4o-transcribe STT, hybrid retrieval with reranking, low-latency TTS.
Live demo
Q&A over PDFs, images, audio, video, and crawled URLs with cited sources. SSRF-defended URL ingest, per-user isolated retrieval, multimodal Voyage embeddings that put text, images and video in one vector space.
Live demo
A public AI demo that spends real money on every click and still cannot be made to overspend: budget reserved before the call and reconciled after, a daily ceiling, a kill switch, and ephemeral provider keys that never reach the browser.
Live demo
URL plus a natural-language prompt becomes validated JSON. Multi-layer SSRF and DNS-rebinding defense, prompt-injection isolation, BYOK across 6 providers, per-key cache scoping.
Live demo
Perplexity-style research in five explicit stages: planning, parallel fetching, source scoring, synthesis that refuses a sentence without a numbered citation, and a self-reflection loop that decides between answering and searching again.
Live demoLanguages and technologies I use
Have a project in mind? Send me a message.
Socials
Find me on social media.
Github
Github @devpedrogomes
LinkedIn
LinkedIn Pedro Gomes
Telegram
Telegram @peugomes