Skip to content
Daniel Wamriew
Projects
Product

AI Powered Marketing Intelligence

A multi-agent marketing application for Revision Base that researches the market, plans campaigns and drafts copy, graphics and video, with a human approval gate before anything is published.

Illustrative
Problem
A small education business needs consistent, well-targeted marketing across Facebook and WhatsApp without a marketing team, but unreviewed AI-generated content is a brand and accuracy risk.
Approach
Six specialized AI agents for research, strategy, copy, design, video and publishing, working from verified product information, with human approval required before any content is published.
My contribution
Direct the AI-assisted development and testing of the application; configure its integrations and evaluate the campaign plans and draft content it produces.

Overview

The marketing intelligence application is an internal tool built to market Revision Base. It brings market research, campaign planning and content production into one workspace, where AI agents prepare the work and a human administrator decides what is published.

Problem addressed

Effective marketing for an education platform means steady, accurate and well-timed content for parents, learners and teachers, tied to real products and seasons such as exams and school holidays. Without a marketing team that is hard to sustain, and fully automated AI content risks publishing claims that are wrong or off-brand.

Technical approach

Six agents, each with a defined job:

Agent Role
Research Gathers market context from the web into research reports with their sources
Strategist Plans campaigns from the product catalog and recent research
Content Drafts weekly Facebook and WhatsApp posts and adds them to the calendar
Designer Produces graphics for approved content
Video Writes scripts and storyboards, and a specification for video rendering
Publishing Checks due content against the approval and scheduling rules

Design principles

  • Verified product information. Campaigns refer to real Revision Base products and prices rather than to anything the models invent.
  • Human approval. Nothing is published until an administrator has approved the exact content, and any later change requires a fresh approval.
  • Safe publishing. Publishing is guarded against duplicates and accidental posts.
  • Video. Approved storyboards can be rendered into short videos.
  • Testing. Automated tests cover the approval and publishing safeguards.

My contributions

I direct the AI-assisted development and testing of the application: setting its goals and safeguards, configuring integrations, and evaluating the campaign plans and draft content the agents produce.

Results

Tests established connectivity with the Revision Base product API and publishing to the Revision Base Facebook page. Scheduled, unattended publishing is still under development, and the system is not yet used for fully automated campaigns.

Lessons learned

  • Human approval is part of the design. Requiring approval of the exact content keeps automation fast without letting unreviewed content through.
  • Be precise about status. A feature that exists in code is not the same as one proven end to end; the project tracks those separately.