What I Built

A guided strategy articulation tool for technology executives. Users answered adaptive questions across domains (governance, enablement, cybersecurity, data privacy), and the system generated board-ready presentations using proven and popular storytelling frameworks (McKinsey/SCR, Duarte, Minto Pyramid, Campbell, etc).

The engine evaluated answer quality with LLM-powered heuristics, tracked progress across strategy dimensions, and exported branded PDF and PPTX decks with watermarks and custom color systems.

Built with Next.js 15, React 19, Supabase, and the Anthropic Claude SDK.

What Killed This Project

In February 2026, Anthropic launched Claude as a native Microsoft PowerPoint add-in. It reads slide masters, layouts, fonts, and brand systems directly inside PowerPoint -- generating structured decks, charts, SmartArt, and process flows without leaving the application.

The output layer that Strategy Storybuilder produced became achievable through a general-purpose tool embedded in the user's existing workflow. Savvy prompters could match the core functionality using a native LLM connector.

The differentiation I was building (curated storytelling formats paired with expert per-strategy-type question trees that shaped the narrative arc) added genuine structure to the thinking process. But for most users, that structure wouldn't have justified the switching cost of adopting a separate tool when a capable LLM was already inside their deck.

What Might Live On

The most durable piece of this project was the question tree engine. Each strategy type (governance, enablement, cybersecurity, data privacy) had a tailored question tree where follow-up paths adapted based on prior answers. The LLM evaluated the quality of each response in real time probing for specificity, flagging vague answers, and surfacing cross-domain connections the user hadn't considered or represented.

That layer wasn't about slides. It was about the strategy itself: forcing rigor into how executives articulate their position, pressure-testing assumptions before they reach a deck. The question trees, the evaluation heuristics, and the cross-referencing logic are patterns worth carrying forward -- whether inside a presentation tool or not.

What This Demonstrated

• Adaptive Q&A engine with LLM evaluation and synonym matching
• SSE streaming generation pipeline with robust JSON parsing
• Supabase auth with RLS, optimistic locking, version conflict handling
• PDF/PPTX export with per-project branding and watermarks
• Full Jira API integration for feature request tracking