Discovery Engine: Self-Service Client Discovery

A Tool for Discovery

Client discovery takes time. You schedule a call, ask questions, take notes, then write everything up into a document. It’s necessary work, but I started wondering: what if a tool could help a customer do that discovery on their own?

Discovery Engine is that tool. It guides clients through structured conversations and produces requirements documents at the end. Instead of scheduling a call, you send them a link. They answer questions at their own pace, and when they’re done, both of you have something concrete to work from.

Bearings

The system uses configuration files called bearings. The name comes from navigation - when you’re traveling somewhere unfamiliar, you get your bearings first. You figure out where you are, where you’re headed, and what’s between here and there. That’s what these discovery conversations do for a project.

Each bearing is a YAML file that defines the discovery flow: phases with starter questions, follow-up probes, and completion criteria. The system (powered by Claude from Anthropic) doesn’t just read from a script. It adapts to answers, asks clarifying questions, and decides when each phase is genuinely complete before moving on.

A typical bearing might have five or six phases: understanding the problem, exploring the current state, defining success metrics, identifying constraints, and scoping an MVP. The conversation moves through these phases naturally, capturing insights along the way.

Clients can pause anytime and resume later. The engine picks up exactly where they left off.

The Technical Stack

Built with Next.js 14, Supabase for persistence, and the Anthropic SDK. Discovery conversations run on Claude Haiku at roughly $0.008 per session.

The architecture prioritizes simplicity. YAML configurations mean non-engineers can modify discovery flows. Real API tests (not mocks) catch integration issues early. Everything deploys to Netlify.

Why Build This

Customers think more deeply when they’re not on a call. They can take breaks, consult colleagues, and return with better answers. The resulting documents tend to be more thorough than meeting notes.

For consultants and agencies, it’s a way to encode discovery expertise once and reuse it across projects.

The tool is live at discovery.paz.land and the source is available on GitHub.