How I'd build Solar AI
I'd reach for Next.js on the frontend, Postgres on the backend with Supabase for auth and real-time features, Stripe for billing, and Twilio for SMS delivery to solar prospects. On the data side, I'd integrate with SEIA member directories and common solar CRMs like JobNimbus via their APIs. The rough estimate is 320-400 hours to get from zero to a production-ready MVP with multi-tenant accounts, email and SMS campaign logic, and basic analytics. That's 4-5 weeks of focused work.
Day-by-day plan
- Day 1-2: Spin up Next.js project, Postgres schema for multi-tenant accounts, Supabase auth, and role-based access control. Set up CI/CD with Vercel.
- Day 3: Build the core prospect model, import logic for CSV uploads, and validation rules for duplicate detection across solar companies.
- Day 4-5: Wire Stripe billing for three pricing tiers (Starter $250/mo, Pro $350/mo, Enterprise custom). Handle downgrades, proration, and failed payment recovery.
- Day 6-7: Build the email campaign builder UI. Template management, prospect segmentation by company size and installer type, send scheduling via Twilio SendGrid.
- Day 8-9: Implement attribution tracking via unique landing page URLs and UTM capture. Build basic Postgres views to power the ROI dashboard.
- Day 10: Set up job queue with Bull and Redis for async email delivery, scheduled triggers, and rate-limiting to avoid spam complaints.
- Day 11-12: Build customer onboarding flow with in-app tutorial, sample campaign templates, and Loom video embeds.
- Day 13: Create admin dashboard for support and churn analysis. Deploy to production and test against live Stripe and Twilio sandboxes.
What's hard about this build
The hardest part is attribution. Solar sales cycles run 3-8 weeks with multiple touchpoints. A demo might close three weeks after an email touch, after five LinkedIn touches, after a phone call. If a customer cancels at renewal claiming they can't prove ROI, you're left guessing whether the tool drove that close. I'd build a UTM-based attribution layer and a self-reported outcome field, but I can't solve this cleanly without customers manually tagging which deals they closed using the tool. Second risk is data quality from SEIA directories and JobNimbus. Stale prospect lists and mismatched company records tank deliverability and erode trust fast. I'd validate every import against phone and domain checks. Third gotcha is churn from IRA policy uncertainty. If federal solar incentives roll back, installer pipelines freeze and customers drop. That's not a technical problem I fix, but it means tight GTM and proof of ROI within 60 days.
What's fast because of AI
Claude accelerates this build in three concrete ways. First, scaffolding: I'd use Claude to generate the multi-tenant schema, Stripe webhook handlers, and email template components, then refine them rather than writing from scratch. That saves 40-60 hours. Second, test coverage: I'd ask Claude to enumerate edge cases for billing and campaign logic, then generate parametric test suites. That compresses days of manual thinking into hours. Third, copy and UX: Claude writes onboarding copy, error messages, and success notifications faster than I'd draft them, freeing me to focus on dashboard mechanics and APIs. I'd also lean on Claude for debugging Twilio or Stripe integration issues; describing the error often surfaces the fix faster than reading docs.
How I'd hand it off
I'd record a 30-minute Loom walkthrough of the admin panel, campaign builder, and billing flows. I'd document the tech stack, deployment process, and Postgres schema in a Notion runbook with guides for common support issues: refunding charges, manually triggering campaigns, fixing duplicate records. I'd set up a 30-day pager rotation for production incidents and brief the buyer on alerting thresholds and slow-query logs. I'd transfer Stripe, Twilio, and Supabase credentials via 1Password and show them how to rotate. I'd be on-call for three weeks post-launch to debug live issues and coach iteration from there.