Podcast Builder

Podcast Builder produces short (10–12 minute) NPR-style conversations among multiple speakers as an alternative entry point into material faculty and students would otherwise only get as a written pre-read. A general-purpose AI model handles source analysis, pedagogy framing, script architecture, and production editing; ElevenLabs (Studio, Text-to-Dialogue API, hosted MCP) handles voice casting, multi-speaker synthesis, and export. The faculty-facing workflow stays simple — advanced model in, finished audio out — while the automation layer this project builds does the coordination between the two.

The pattern originates in the cloned-voice podcast pipeline built for the Legal Education at Model Velocity microsite, and generalizes it into something repeatable rather than bespoke per course.

What we’re doing

• Build the API/automation layer that drives ElevenLabs Studio and Text-to-Dialogue from a locked script and a small voice-casting spec, without faculty needing to touch MCP or API calls directly.
• Standardize the production inputs an episode needs: script, voice bible/episode plan, production and performance notes, casting criteria.
• Pilot the workflow on a real episode rather than a demo: an alternative audio entry point into the pre-read for the 2026 faculty retreat’s AI session.
• Generalize what works into a template other courses and programs can reuse.

Why this matters

A well-produced pre-read still asks a lot of a reader’s time before they’ve decided the topic is worth it. A short, well-produced conversation lowers that barrier and can carry nuance — disagreement, qualification, someone changing their mind mid-argument — that a summary flattens out. Making that format cheap enough to produce per-course, rather than reserved for flagship microsites, is what turns it from a one-off into infrastructure.

Status

Planning. Workflow being built alongside its first real pilot: an episode accompanying the September 2026 faculty retreat’s AI pre-read.