How I'd build Sealwax
I'd reach for Next.js on Vercel for the frontend, Postgres on Railway for the database, Stripe for billing, and DocuSign's eSignature API for the signing workflow. I'd use Clerk for auth, Resend for email, and Twilio for reminder SMS. The document generation layer would be template-based with clause variations keyed to the agent's clarifying questions. Rough estimate: 50-55 billable hours to get to a launchable MVP, maybe 40 if the initial feedback loop is tight.
Day-by-day plan
Day 1: Provision Postgres schema (users, accounts, draft NDAs, signatures), Clerk auth setup, Next.js scaffolding with API routes, tenant isolation middleware. Day 2: Build the intake form and conversational clarifying-questions flow. Day 3: Wire clause templates and document generation logic; test output for the five most common jurisdiction/engagement combinations. Day 4: Integrate DocuSign API for draft review, signing links, and webhook handlers for signature events. Day 5: Stripe product setup across three tiers, billing portal, webhook handlers for subscription state. Day 6: Email notification pipeline (Resend), 24/72-hour reminder jobs using cron, signed PDF archival to Postgres bytea. Day 7-8: QA, landing page copy, Product Hunt launch prep, Vercel and Railway deployments.
What's hard about this build
The DocuSign integration is the technical spine, and webhook reliability matters: a dropped signature event means a party thinks they've signed but the system never recorded it. The second hard thing is that one poorly-drafted clause ruins trust instantly. The agent needs to flag jurisdictional gaps (California non-compete bans, New York consideration gaps) and know when to refuse to generate a document at all, because a template-generated NDA that fails in court becomes a legal and PR liability overnight. Third, the free tier will attract price-sensitive users who will never convert to paid. I'd need to be disciplined about not over-serving the free tier or the paid tier becomes a technical afterthought. Fourth, multi-party routing creates state-management complexity: an unilateral NDA that names three signing parties in different orders introduces edge cases that tests need to catch.
What's fast because of AI
Document generation scaffolding compresses a full day: Claude generates the base template structure and clause variations in minutes, then I validate and refine. The clarifying-questions flow takes another half-day instead of a full day because Claude enumerates the edge cases I'd miss (IP carve-outs for pre-existing work, competitive restrictions embedded in confidentiality, perpetual language courts reject). Test coverage that would normally be an afterthought gets written in parallel because Claude generates realistic test vectors for each jurisdiction. Copy for the product UI, the email templates, and the error messages gets drafted and iterated in an evening instead of bouncing back and forth. Debugging the webhook payload mismatch with DocuSign normally eats four hours; Claude shortens it to one by helping me reason through the state machine and spot the missing ACK step.
How I'd hand it off
I'd record a Loom walkthrough of the admin dashboard, the clarifying-questions flow, and the signature tracking interface. I'd leave a runbook covering the Stripe and DocuSign API key rotation, the cron job schedule, and how to interpret signature-state logs when a user reports a missing document. Thirty-day pager rotation would mean you or a designated ops person getting a Slack alert if a webhook fails three times in a row. I'd transfer Postgres credentials to your vault, create a Vercel team for deploys, and leave a Linear board with the next ten features ranked by monthly-active-user impact. First call would be a handoff walkthrough; first month of questions I'd answer in a Slack thread or a short call.