Skip to content
Keith McAleer
All work

Agentic marketing workspace · Designed around a university marketing team

Venti

Making it easy for a whole community to contribute to its own marketing.

Status

In development · not yet deployed

Built with
Next.js, TypeScript, PostgreSQL / Prisma, Claude Agent SDK, HubSpot API

Venti is not deployed yet. This page describes the system as built so far and what it's for, not a launched product.

Ventiin development

Agents

  • EurusStrategy lead
  • AuraBrand
  • ZephyrusEmail campaigns
  • AeolusMarketing loop & leads
  • BoreasContacts
  • NotusPages & forms
  • AellaVideo

Tools by area

  • Campaigns9
  • Brand6
  • Contacts6
  • Forms & pages6
  • Video5
  • Strategy & reports5
  • Leads4
  • Agent hand-offs4
  • Lists & utilities11
Diagram of the system as built so far, from its agent and tool registry. Not a product screenshot.

A multi-agent workspace that combines structured contribution forms, reusable brand context and agentic workflows. Faculty, staff and students can hand over stories and campaign inputs with little effort, and a small marketing team can turn them into useful content.

Problem

A university entrepreneurship center has an enormous amount of expertise spread across faculty, staff, students, programs and partners, and a marketing team of a handful of people. The constraint isn't ideas. It's that contributing to marketing takes too much effort: a story needs a meeting, a draft, rounds of edits and someone who knows the brand.

So most institutional knowledge never becomes content.

Insight

This is an operating-system problem more than an automation problem. Content doesn't scale by generating more of it. It scales when contributing gets easier, so the people with the knowledge actually share it.

  1. Make contribution easier
  2. Capture more institutional knowledge
  3. Create more useful content
  4. Increase reach

What I built

A workspace of seven specialized agents. Each covers a different area (strategy, brand, email campaigns, leads, contacts, pages and forms, and video) and runs on the model that suits it. The agents share a scratch space to hand work to one another, and a lead agent recommends the next step.

Brand is represented as structured data rather than a PDF of guidelines, so every agent writes in the same voice and follows the same rules.

  • 50+ tools across brand, campaigns, leads, contacts, lists, forms, pages, strategy and reporting
  • Structured forms that turn a contributor's answers into campaign-ready inputs
  • A HubSpot integration that reads contacts and writes lead scores back to the CRM
  • Automated tests from the start

My role

Product definition, agent and tool design, the brand data model, and the build itself.

Evidence

  • In development, not yet deployed
  • 7 agents and 50+ tools implemented so far
  • HubSpot integration that reads contacts and writes back lead scores

What it shows

  • Treating marketing capacity as a systems problem
  • Multi-agent design applied to a real organizational constraint
  • Clear scoping: what it does today versus what it's for

I'm interested in consequential work at the intersection of AI, product and growth.

I'm based in San Francisco and talking with teams working on how AI changes products, marketing and knowledge work.