SermonAI

Pastors and church staff spend hours every week turning one sermon into summaries, discussion questions, devotionals, blog posts, and verse lists — or wrestling with generic AI chat that drifts off their message. The product had to start from messy real inputs (recorded video or a pasted manuscript), keep outputs grounded in that source so the voice stays the pastor's, and make the path simpler than an open-ended LLM chat. Success meant a weekly workflow people could finish: generate what they need, edit in place, then copy or download — not a black-box that invents a sermon they didn't preach.

I designed around the pastor's source material, not a blank prompt. Intake is video or manuscript; users pick content types (summary, verses, questions, blog post, devotionals), then run a staged pipeline with a loading state per step so progress is clear to the user. Guardrails keep generation tied to the transcript or manuscript so the model is far less likely to invent beyond what was said. Results open in the dashboard for edit, regenerate, copy, and download. An early one-spinner build hid stalls and broke trust, so I split the job into visible steps. Faithfulness to their voice was day-one. I use Cursor to ship faster while owning architecture and judgment.

The dashboard is the product: upload or paste a sermon, choose content types, watch each stage complete, then edit in place before copy or download. No social publish or posting integrations — export stays intentional. Team workspaces and roles support church staff; Stripe covers trial and subscription. Next.js App Router monorepo, Supabase, Deepgram, and OpenAI structured outputs sit underneath; the design center is a grounded, editable pipeline simpler than an LLM chat for this weekly job. The marketing site describes the same workflow the app delivers.

Live at sermonai.pro for ministry teams on trial and subscription paths. Users go from sermon or manuscript to multiple editable content types in one dashboard — regenerate and revise until it matches their voice, then copy or download. I own research, flows, API architecture, billing, and iteration, including where Cursor accelerates delivery. The outcome I care about: AI as a tool under their control, grounded in their content, easier than a generic chat interface for this work.