A multi-agent AI marketing-automation platform

Where things stood.

Managing paid media well is relentless, manual work — research, copy, QA, reporting, follow-up — and it doesn't scale without more hands. As a Computer Science engineer who happens to run campaigns for a living, Sunny set out to break that ceiling: build a real platform where a fleet of AI agents handles the repetitive execution and the human sets strategy. The constraint that made it interesting — build the whole thing himself, orchestrating agentic AI as the engineering team.

How it got fixed.

1. 01

   Design the multi-agent architecture

   Broke marketing execution into discrete jobs — research, content, QA, segmentation, reporting — and assigned each to a specialised agent, with 20+ running concurrently and a human approving the decisions that matter.

2. 02

   Run the agents as Node.js services

   Orchestrated the fleet as Node.js background services so tasks run continuously and reliably off a queue — not as brittle one-off scripts — routing each job to the model that does it best (Claude, ChatGPT, NVIDIA APIs, Google AI Studio, Ollama/DeepSeek).

3. 03

   Automate content with a QA gate

   Built content-generation pipelines with an automated QA pass in front of anything that ships, so volume never came at the cost of quality or brand safety.

4. 04

   Wire a custom CRM with messaging built in

   Shipped a custom CRM with automated email and WhatsApp (WATI) follow-up, so a lead the agents generate is captured, nurtured and routed without manual handoff.