Ellum AI

The problem

Small marketing teams drown in the gap between strategy and execution. They know what they want to say; they do not have the hours to plan a calendar, draft for six platforms, and publish on schedule. Generic AI writing tools made the drafting marginally faster and left every other part of the job untouched.

What I did

• Owned the product from discovery through MVP and into a V2 architecture defined across eleven milestones and seven epics.
• Re-architected the content engine from single-shot prompting to an orchestrator model that routes work to specialised agents, skills and tools, and reports on drift across the fleet.
• Designed an AI-enabled onboarding flow that captures brand identity, goals, tone and audience as structured context every downstream agent can use.
• Specified the social infrastructure rewrite covering LinkedIn, TikTok and YouTube integrations with per-user resource allocation to dodge shared API rate limits.
• Defined DeepEllum, the catalog customers hire from: pre-built specialists installed in one click, custom AI Staff provisioned with their own title, department and brief, and subscription-aware installation so entitlements match what a plan pays for.
• Made the work auditable rather than a black box: every output grounded in the Knowledge Center with citations, a network view of each handoff, and an inbox that holds anything needing approval before it goes live.
• Ran the pilot with roughly 50 users, converted feedback into prioritised iterations, and produced the investor-ready materials that supported fundraising.
• Led a distributed team of eight across backend, frontend, design, QA and marketing on a weekly standup and sprint cadence.

Where it landed

Stack & tools