Scaling to 4 Concurrent AI Programs and Architecting the Platform to Unify Them

Situation

A large public institution had one AI initiative and wanted four: healthcare ePortfolio, interactive math, AI tutoring, financial training. Each had its own stakeholders, timeline, and definition of done. Run them as separate projects and you get separate architectures, separate vendor relationships, and a maintenance bill nobody signed up for. Nobody had committed to a shared foundation yet.

What I Built

• Scaled from 1 to 4 concurrent AI programs, each at a distinct delivery phase, without collapsing them into a single undifferentiated backlog or losing any program's specific stakeholder relationships.
• Ran a 20+ vendor evaluation, benchmarked against systems-of-record data, and delivered a weighted recommendation to senior leadership that set platform selection and a six-capability roadmap.
• Led SaaS vendor engagement from contract through pilot launch, coordinating LTI 1.3 integration across internal and third-party platform partners and delivering five artifacts to senior leadership within 30 days of kickoff.
• Prototyped in working software rather than specifying and handing off, building daily with Claude Code and agentic coding tools, so architecture and UX decisions were tested against something running before they reached a document.
• Built a GenAI-powered workflow automation system for my own team, seven purpose-built AI skills orchestrating retrospectives, documentation, research synthesis, and tracking across four active workstreams.
• Architected the full product spec for an institutional AI platform serving more than 200,000 learners, across eight-plus domains: RBAC, LTI 1.3/SSO, data hierarchy, conversation design, and UX standards, from prototype through architecture-lock.

The Hardest Problem

Getting four programs, four separate stakeholder groups, to agree to build on one shared architecture before any single one of them needed it badly enough to fund it alone. The case had to run on future maintenance cost and integration risk. That's a much harder sell than "this feature ships faster." The vendor evaluation data made it concrete, not theoretical.

Outcome

The platform spec reached architecture-lock across all eight-plus domains. One vendor selection and roadmap, and all four programs build against it. The team's own workflow automation system runs in parallel, applying the same AI-native practices internally that the platform is meant to enable institution-wide.

Skills Demonstrated

AI platform strategy, vendor evaluation and selection, enterprise architecture across concurrent programs, GenAI workflow tooling, cross-stakeholder alignment at scale.