Skip to content
Bryan Silva

Bryan Silva

AI & AUTOMATION

00Trigger

About

AI & GTM Engineer | Automation Architect

Belo Horizonte, Brazil · Remote

I'm Bryan Silva, an AI & GTM engineer based in Belo Horizonte, Brazil. I design and ship AI and automation systems to production for marketing agencies and B2B clients. I'm passionate about solving problems: I like to understand the operation in depth before drawing the flow, and to deliver systems the team can use and maintain without me.

n8n automations, API integrations, data platforms and AI agents applied to sales, marketing and operations. With guardrails, human-in-the-loop and AI cost under control.

See the case studiesGet in touch
TriggerProcessAIToolsOutput

AI and automation engineering for teams that need to scale.

Experience

  • AI Engineer & Automation Architect

    Marketing agencies · Jan 2026 – present

  • Automation Engineer

    Freelance · Jun 2025 – present

Languages

Portuguese (native) · English (professional) · Spanish (professional)

Beyond the code

  • I turn business problems into systems

    I start from the routine of the people running the operation, whether sales, marketing or support, not from the tool.

  • Logic before the model

    My analytical side decides: rules where the problem is deterministic, AI where there's interpretation.

  • Curious and a fast learner

    I'm always researching technologies and better ways to solve a problem, and I put what I learn into practice on my own initiative.

  • Communication and teamwork

    Both ends of the work depend on it: at the start, I listen to the team to understand the problem; at the end, I teach and enable the people who will use the systems, automations and agents.

  • I take it from design to operations

    Architecture, delivery and follow-up in production, end to end.

  • I build for scale

    No templates: every system is designed to grow with the operation, with configuration in data and documentation for the team.

01Input

Results from systems in production

Each row is a confirmed number, next to the system that produced it.

ValueMetric
2,000+contacts handled per month, automatically
−60%manual work in prospecting
+45%in the quality of worked leads
50+ hfreed up per month in operations
30+automated processes
70%faster processing
02Process

Systems in production

  1. 01AI Agent · WhatsApp · RAG

    WhatsApp prospecting with an AI agent

    Finds businesses on Google Maps, makes first contact, and an AI agent answers, qualifies and hands the ready lead over to the team.

    Technical decisions

    • RAG on pgvector with one rule: it only answers from what it found.
    • Human-in-the-loop: if someone from the team writes in the chat, the AI pauses itself.
    • Redis debounce for split messages (~22 s window) before replying.
    • Multimodal: text, audio (Whisper), image and document.

    Stack

    • n8n
    • Evolution API
    • Supabase
    • pgvector
    • Redis
    • Postgres
    • Serper
    • OpenAI
    • Whisper
    • Gemini
    • Google Sheets
    View full case study
    ai-sales-agent

    1. 01

      Lead sources

    2. 02

      Shared database

    3. 03

      First contact

    4. 04

      Message buffer

    5. 05

      RAG agent

    6. 06

      Human handoff

    contacts/month
    2,000+
    manual work
    −60%
    lead sources
    3

  2. 02GTM Engineering · Clay · LLM guardrails

    Email prospecting with Clay, AI scoring and a guardrail

    Every lead enriched in Clay gets a score, a tier and, when it's worth it, a personalized email checked against facts.

    Technical decisions

    • Strict split: facts in deterministic code, interpretation in the LLM.
    • Rule-based base score 0–40; the LLM normalizes to 0–100 and adjusts by ±25 at most.
    • Evidence outside the closed list of facts sends the lead to manual review.
    • Status lock before sending: no duplicate emails.

    Stack

    • n8n
    • Clay
    • OpenAI
    • Supabase
    • Gmail API
    • Webhooks
    View full case study
    clay-ai-scoring

    1. 01

      Clay webhook

    2. 02

      Signals → facts

    3. 03

      Rule-based score

    4. 04

      LLM analysis

    5. 05

      AI copy

    6. 06

      Guardrail

    7. 07

      Locked sending

    lead quality
    +45%
    tiers (A–D)
    4
    emails/day
    500+

  3. 03AI for data · MCP · Full-stack

    AI-powered paid media operations: platform and MCP agent

    A platform and an agent so the team sees every account in one place, produces standard reports and gets alerts.

    Technical decisions

    • Read-only by default: any change to an account needs approval.
    • Controlled AI cost: Haiku for chat, Sonnet for analyses, prompt caching and a budget per workspace.
    • Ads × GA4 check: missing data never becomes “zero results”.
    • AES-256 credentials and multi-workspace with roles.

    Stack

    • React
    • TypeScript
    • Node.js
    • MySQL
    • Claude
    • MCP
    • Google Ads API
    • GA4
    • Meta Ads
    • TikTok Ads
    • LinkedIn Ads
    • n8n
    • WhatsApp
    • PM2
    • Nginx
    View full case study
    ai-media-ops

    1. 01

      Media APIs

    2. 02

      Daily sync

    3. 03

      Normalized table

    4. 04

      Platform

    5. 05

      Monitoring

    6. 06

      MCP agent

    freed up/month
    50+ h
    platforms in one table
    5 → 1
    to set up a manager
    ~15 min

  4. 04Agentic architecture · Human-in-the-loop

    Multi-agent personal assistant

    A WhatsApp assistant that handles calendar, tasks and meeting actions, secure by construction.

    Technical decisions

    • The agent never talks to external APIs, only to sub-workflows; credentials live in the tool.
    • Destructive actions don't exist in the graph without a confirmation step.
    • The sender ignores the destination the LLM asks for and only uses the one on the client record.
    • Atomic claim and 3-layer dedupe; an ambiguous time becomes a task, never an event.

    Stack

    • n8n
    • OpenAI
    • Gemini
    • Whisper
    • Postgres
    • Redis
    • Google Calendar
    • Google Drive
    • Evolution API
    • Notion
    View full case study
    multi-agent-assistant

    1. 01

      WhatsApp

    2. 02

      Central agent

    3. 03

      Sub-workflows

    4. 04

      Crons and transcripts

    5. 05

      Human confirmation

    6. 06

      Calendar and tasks

    workflows
    8
    tables
    7
    of crons in < 1 s
    ~95%

  5. 05AI for data · Paid media

    Campaign monitoring with an AI analyst

    Code decides whether there's a problem; the AI only explains and prioritizes, and the alert lands on WhatsApp.

    Technical decisions

    • Code decides whether there's a problem; the AI only explains and prioritizes.
    • Numbers come from a calculator tool, never estimated by the model.
    • Rules per campaign type: Search, PMax and everything else.
    • A Meta Ads variant catches active campaigns with no delivery.

    Stack

    • n8n
    • Google Ads API
    • Meta Graph API
    • OpenAI
    • Google Sheets
    • Evolution API
    View full case study
    campaign-monitoring

    1. 01

      Weekly trigger

    2. 02

      Campaign data

    3. 03

      Rules by type

    4. 04

      Healthy

    5. 05

      AI analyst

    6. 06

      WhatsApp alert

    clients
    10+
    steps
    6
    step where healthy campaigns stop
    3
03AI

How I build

I build with the whole process in mind, from the moment data comes in to the routine of the people who use it, so every system reaches production safely and ready to scale.

04Tools

Tools I work with

Orchestration, LLMs, GTM, CRM and data: where automation meets the team's daily work.

  • Automation

    The core of the automations: workflows and sub-workflows, AI agents, API and webhook integrations, with state in a database and error handling.

    • n8n
    • Make

  • LLMs

    OpenAI, Claude and Gemini in agents, classification, extraction and analysis, with the right model for each task.

    • OpenAI
    • Claude
    • Gemini

  • SalesTech

    Enrichment and waterfalls, intent signals and Clay integrations with CRMs and n8n.

    • Clay
    • Apollo
    • n8n
    Case study in preparation

  • CRMs and ops

    Pipelines, lead routing, two-way sync and data hygiene in CRMs, plus operations automations in project management tools.

    • HubSpot
    • Pipedrive
    • RD Station
    • monday
    • Notion
    • Asana
    Case study in preparation

  • AI for data analysis

    Power BI and Looker Studio dashboards and LLM analyses over business data.

    • Power BI
    • Looker Studio
    • LLM
    Case study in preparation

Full stack

Automation

  • n8n (advanced)
  • Make
  • Zapier
  • Webhooks and REST APIs
  • Evolution API

AI

  • OpenAI
  • Claude / Claude Code
  • Google Gemini
  • RAG and vector databases
  • Agents and MCP
  • Microsoft Azure AI
  • Google Vertex AI
  • Whisper

Data

  • SQL
  • Supabase / Postgres
  • pgvector
  • MySQL
  • Redis
  • Google Sheets
  • Excel

GTM, CRM and Ops

  • Clay
  • Apollo
  • HubSpot
  • Pipedrive
  • RD Station
  • CRMs in general
  • monday
  • Notion
  • Asana
  • Gmail API
  • Outlook

BI and Marketing

  • Power BI
  • Looker Studio
  • Google Ads
  • GA4
  • Tag Manager
  • Merchant Center
  • Meta Ads
  • TikTok Ads
  • LinkedIn Ads

Development

  • React
  • TypeScript
  • Node.js
  • Python
  • PM2
  • Nginx
05Branches

Other systems

Other automations running in parallel for operations, media, data and knowledge.

  • Editorial calendar monitor

    Reads a loose document, detects status changes and alerts the right person.

    • n8n
    • Google Docs
    • Supabase
    • WhatsApp

  • Media budget pacing

    Compares each account's spend with its budget and projects overspend before it happens.

    • n8n
    • Google Ads API
    • Google Sheets

  • Data for BI dashboards

    Consolidates and organizes data as the source for Power BI dashboards.

    • n8n
    • Google Sheets
    • Power BI

  • AI analysis on top of the CRM

    Connects the CRM to an external database, where an AI agent runs a specific analysis over the sales data.

    • n8n
    • CRM
    • SQL
    • LLM

  • Prospecting follow-up

    Tracks prospecting emails with a state machine and schedules each follow-up.

    • n8n
    • Gmail

  • Daily AI digest

    Summarizes technical content every morning and emails it to the team.

    • n8n
    • RSS
    • OpenAI
    • Gmail
06Output

Let's build something amazing together.

Want to see one of these systems running? I can show you the workflow, a real execution and the result.