How I'd build Compliance Training AI
I'd use Next.js for the customer-facing app and course delivery, FastAPI for the backend API, and Postgres as the data layer. Stripe handles billing and subscription management, Resend sends transactional and onboarding emails, and Cloudinary stores and streams course videos. I'm estimating 140-160 hours total: around four weeks at 35-40 hours per week, which comes out to roughly $10,500-12,000 in labor; the remaining $20,000 of the $32,000 total covers Stripe fees, cloud hosting, initial legal review of compliance templates, and buffer for the harder technical and content problems.
Day-by-day plan
Day 1-2: Provision Next.js monorepo, Postgres schema for users, tenants, courses, lessons, and enrollment records. Implement NextAuth with email and social login. Day 3: Stripe integration for the three pricing tiers (Basic $99, Pro $129, Enterprise $149 monthly). Day 4-5: Build course delivery UI with video player integration, lesson completion tracking, and embedded quiz submission. Day 6: Admin dashboard for customers to upload course materials, edit descriptions, and manage user seat assignments. Day 7: Set up Resend email onboarding sequence (welcome email, day-3 engagement reminder, week-1 upsell check). Week 2: Implement completion certificates, state-specific course variant selection logic, and strict tenant data isolation tests. Week 2-3: Analytics dashboard for customer admins, mobile responsiveness, and performance optimization. Week 3-4: Load testing, deployment to Vercel, pager runbook creation, and credentials handoff.
What's hard about this build
The core technical risk is content liability. State employment law on harassment training, HIPAA standards for healthcare training, and OSHA safety rules all drift by jurisdiction and update annually. If I generate California supervisor harassment training and a customer applies it in Texas, or if Claude's synthesis of OSHA 10 misses a recent rule change, the customer gets audited and names me as the vendor. I need to build an audit trail (who reviewed, when, what source material, what version) and flag every jurisdiction-specific claim for human legal review before shipping to a customer. That's not just code; it's a lightweight CMS workflow with approval gates. The second risk is data isolation: a tenant's courses and completion records must never bleed to another tenant in queries, which means careful schema design, row-level security rules, and auditing every join. The third is concurrent video uploads and streaming, which Cloudinary mostly solves, but concurrent course completions need database-level guards to avoid double-counting.
What's fast because of AI
Claude collapses the course scaffolding phase from days to hours. I describe the course structure once (modules, lessons, quizzes, completion logic, state variants), and Claude generates the Postgres migrations, FastAPI endpoints, Next.js components, and test suites in a single session. Copywriting for product UI, emails, and error messages that I'd normally spend a day on takes two hours: I specify tone and audience, Claude generates five variations, I pick the best and refine. AI also accelerates edge-case enumeration: I ask Claude to list every way a user can corrupt their own course data or break a lesson flow, then build guards against those cases faster than I'd catch them manually. Debugging the multi-tenant query logic is faster too: I paste a schema diagram and a suspicious query, Claude spots potential data leaks. I'd estimate AI cuts scripting and testing work in half and copywriting by 60%.
How I'd hand it off
I'd create a Loom walkthrough of the admin dashboard: uploading a course, selecting state-specific content, and auditing change history. I'd write a runbook covering Stripe webhook failures, deployment procedures, and how to respond to a customer reporting a course error (open a GitHub issue, pull request to add a disclaimer or content fix, code review, merge). I'd commit to a 30-day rotating pager where I'm on call one week a month for bug reports and state-law updates. Source code, database credentials, and Stripe API keys go into a shared 1Password vault with emergency-access procedures documented. The Linear board and GitHub repo are yours; I'll hand over all recent decisions and architectural notes in a Google Doc.