sampaiodev.
PTEN

Gabriel Sampaio · sampaiodev

APIs, automation and applied AI that become running systems.

I turn manual operations into production-ready systems with APIs, queues, AI agents, databases, logs and deploys. From input to useful output, with errors handled and infrastructure built to keep running.

View projectsTalk about a system

Or follow the work:GitHubLinkedInBlog

  • 6selected projects
  • 4open-source
  • 1flagship

StackPython · FastAPI · n8n · Docker · PostgreSQL · Redis · Applied AI

Proof before promise

Three flows you can open and inspect.

Each case connects a real problem to publicly verifiable code, documentation and operations. No invented client, assumed scale or unsourced number.

Flagship

meet-transcription

Problem

Meeting recordings land in Drive, but turning every MP4 into useful text requires downloading files, calling a provider, handling queues, persisting results and delivering TXT without blocking the UI.

Verified result

The repository shows web, worker, migrations, a Redis queue and PostgreSQL as the source of truth. Compose documents startup order and keeps heavy transcription outside the request.

Flow diagram
  1. InputGoogle Drive
  2. APIFastAPI web
  3. Queue/lockRedis
  4. ProcessingWorker
  5. TranscriptionDeepgram or local Whisper
  6. StoragePostgreSQL
  7. OutputTXT
Open-source

notebooklm-api

Problem

NotebookLM is useful in the UI, but repeated flows need an API, account selection, session checks and clear limits when Google asks for a challenge.

Verified result

The current public version documents default-deny authentication, per-account session isolation, scoped jobs and artifacts, and tests for security invariants.

Flow diagram
  1. InputREST/CLI/UI client
  2. APIFastAPI
  3. SelectionAccount
  4. Sessionstorage_state
  5. AutomationNotebookLM/browser
  6. OutputHTTP/API response
Open-source

dashboard-sheets

Problem

Spreadsheets become manual operations when nobody turns the data into a screen. Deployment also needs to avoid credentials inside the image and run on AMD64/ARM64.

Verified result

Compose runs with a non-root user, read-only filesystem, dropped capabilities and a healthcheck. The workflow publishes AMD64 and ARM64 images to GHCR.

Flow diagram
  1. InputGoogle Sheets
  2. AccessCredentials/API
  3. ProcessingStreamlit
  4. OutputWeb dashboard
  5. OperationsDocker/GHCR

Featured projects

Systems, not just repositories

Each project starts with an operational problem and shows the flow that turns input into a useful result.

View all projects

About me

I like the part where the idea meets operations

Gabriel SampaioPython · automation · applied AI · infrastructure

I understand the process, design the flow, implement the integrations and handle the parts that keep the system alive: queues, logs, healthchecks, handled errors and deploys. I'm starting an ADS degree while deepening that foundation through open-source projects and products of my own.

About me

How I work

  • Full flow: input, processing, queue, handled errors, persistence and output.
  • Real infra that boots, restarts and is versioned — it doesn't run only on my machine.
  • Simple observability: useful logs, healthchecks and no sensitive data leaking.
  • Applied AI for real: agents and RAG wired to operations, not a demo.

Trajectory

How I got here

  1. 01Education

    Starting an ADS degree

    A formal foundation in Systems Analysis and Development.

  2. 02Python

    Python and automation

    First projects pulling manual processes out of improvisation with scripts and APIs.

  3. 03Infra

    APIs and containers on a Linux VPS

    Docker, self-hosted deploys and an environment that boots, restarts and is versioned.

  4. 04Pipelines

    Transcription and AI integrations

    meet-transcription, transcricao and flows wiring LLMs to operations.

  5. 05Product

    OrkestraAI and product automations

    n8n workflows productized into an AI-agent SaaS.

  6. 06Next step

    Backend, applied AI and stronger infra

    Deepening architecture, observability and real operations.

Let's talk

Send me the manual process.

I will reply with architecture, scope and the next step — without vague promises or hiding the operational work.

Describe the process
Or follow the work