practically.dev

The problem

Product managers, founders and go-to-market leads work closely with engineers but often can't follow the technical language, which makes roadmap and AI product decisions harder.

What I did

• •Learning tracks with lessons, progress tracking, a reading list and downloadable course materials.
• •A plain-English glossary with inline tooltips, company architecture teardowns and guides to AI tools.
• •An admin course generator that runs planner, builder, critic and refinement agents, with a human approve or revise step for every module.
• •Stripe embedded checkout with a paywall on lessons, plus onboarding and user profiles.
• •A newsletter with admin email campaigns sent on a schedule through Resend, and product analytics in PostHog.

How it works

Next.js App Router with Supabase auth and Prisma on Postgres. Course generation is an explicit state machine that moves from planning through module generation, critique and refinement to review and publishing. Each step is stored as a job with its own status, and every LLM call asks for JSON and records token use.

Decisions and trade-offs

• •AI-written content always passes critique and refinement, then a human gate, before it can be published.
• •Moved from SQLite to Postgres once user progress and course management arrived.
• •Kept Stripe test and live keys split by environment, and excluded webhook routes from middleware so signatures verify against the raw body.

Where it stands

Live in beta.