I build AI products.

What it does: Magpie reads the comment sections under a creator's competitors and turns them into ranked, evidence-backed video ideas. The product bet: the best signal for "what should I make next" isn't keyword volume, it's what real viewers are asking for and complaining about in comments that no one is answering yet.

The product thinking: Most tools stop at shallow semantic matching, "these comments mention X." That surfaces noise, not opportunity. So I built Magpie to reason about a niche the way an insider would, understanding the actual vocabulary, sub-topics, and unmet questions of a space, then cross-referencing them against what competitors already cover to find the genuine gaps. The output isn't "people said this," it's "here's an idea, here's the comment evidence behind it, and here's why it's underserved."

The hard part I'm proud of: a niche-creation agent that handles verticals the system has never seen. When a user submits an unfamiliar niche, the agent calibrates it on the fly, building a structured understanding of that space so the analysis stays accurate instead of generic. Every agent decision is logged, and the system is designed to surface its own failures rather than hide them, because a wrong answer a user can't see is worse than an honest "I couldn't do this well."