Contractor Lead Refund Automation: Recover Angi & HomeAdvisor Credits

AI Prompts to Build This

Copy these build prompts into Claude, Cursor, or your AI coding tool. Create a free account to unlock the full research behind them.

1. 1. Project Setup

   Build the weekend MVP of "Refundra": a contractor uploads 90 days of paid leads with call transcripts and gets a dollar number for leads that look refundable, plus a ready-to-send claim for each one. It does not file anything on the platforms.
   
   Stack: Next.js (App Router), TypeScript, Tailwind, Supabase (Postgres, Row Level Security, Auth with email magic link), the Anthropic API to classify each lead against a platform's refund reasons. Leads come in by CSV with a transcript column. Deploy on Vercel.
   
   Tables (Row Level Security on, scoped to the account that owns the rows):
   - accounts(id, user_id, business_name, trade)
   - platform_rules(id, account_id, platform, window_days, reason_labels jsonb) platform is angi, homeadvisor or bark; the owner fills the values from each platform's own policy page
   - leads(id, account_id, platform, external_lead_id, charged_at, amount_cents, service_type, zip, transcript)
   - classifications(id, lead_id, category, confidence, quotes jsonb, model, prompt_version)
   - claims(id, lead_id, reason_text, window_closes_at, status, credit_cents) status is draft, filed, approved or denied
   
   Screens: /login, /import (CSV), /leads (the dollar number and the list), /claims.
   Env vars (names only): NEXT_PUBLIC_SUPABASE_URL, NEXT_PUBLIC_SUPABASE_ANON_KEY, SUPABASE_SERVICE_ROLE_KEY (server only), ANTHROPIC_API_KEY.
   
   Do not build: billing, plans or a recovery fee, browser automation or Playwright filing, CallRail or Twilio connections, audio transcription, Thumbtack, Google Local Services reconciliation.
   Done when: npm run dev starts, you can sign in, and the five tables exist with Row Level Security on.

2. 2. Core Feature

   Build the one feature that proves Refundra: find the leads that look refundable and draft the claim from the call itself.
   1. Import a CSV with columns platform, external_lead_id, charged_at, amount, service_type, zip, transcript. Show the rows that failed to parse and why.
   2. For each lead, send the transcript and the platform's reason_labels to the Anthropic API with a strict JSON schema { category: "wrong_service" | "out_of_area" | "disconnected" | "duplicate" | "already_hired" | "not_refundable", confidence: number from 0 to 1, quotes: string[] }.
   3. Check every quote in code: it must appear word for word in the transcript. Drop any category that has no valid quote and mark that lead not_refundable.
   4. Work out window_closes_at as charged_at plus the platform's window_days. If window_days is empty, show "Check this platform's policy" instead of a date. Never guess a window.
   5. /leads shows one number at the top: dollars on leads that look refundable with the window still open. Under it, the list with category, confidence and the days left.
   6. For each refundable lead, create a draft claim: the lead's details, the category, the quoted lines and a short reason text in plain words. The contractor copies it into the platform's own form. A status control lets them mark it filed, approved or denied and enter the credit amount.
   7. Show recovered dollars (approved credits) beside the total spent on leads.
   Rules: never promise a cash refund; say credit unless the platform pays cash. Never claim a platform has a dispute API.
   Empty state: with no leads, show the CSV columns and a sample file to download.
   Done when: a 20-lead sample file produces categories with quoted lines, the top number matches the sum of the eligible open-window leads, and marking a claim approved with a credit updates the recovered total.

3. 3. Landing Page

   Build a one-page landing site for Refundra, which finds refundable leads for home-service contractors.
   Hero: "Your Angi and HomeAdvisor credits, found before the window closes." Sub: "Upload 90 days of leads and calls. Get one dollar number and a claim written for each lead." One button: Join the waitlist for the free Refund Audit.
   Sections: a sample audit result card with one big number marked as an example, how it works in four steps (upload, classify, claim, track), a placeholder for a real recovery story to replace with yours, and an FAQ on why you file the claim yourself, credits versus cash and how your call data is handled.
   Waitlist: store the email in a waitlist table in Supabase. No other service.
   Style: Geist, an off-white background, near-black type, one lime accent.
   Done when: the page renders on a phone and a submitted email appears in the waitlist table.

4. 4. Branding Package

   Use a design or image tool for this one. A coding agent cannot draw a logo.
   Brand for Refundra: a wordmark and a small mark that reads as a receipt stamp, not a phone.
   Colors: near-black, one lime chip and two warm grays. Type: Geist, with tabular numbers for dollars.
   Deliverables: wordmark, stamp icon, an audit result card, three sample audit emails that open with one number marked as an example, and a one-page brand sheet with three do-not-say rules: no "AI-powered", no "set and forget", never promise cash when the platform pays credit.
   Done when: each deliverable is saved in one folder and the stamp is still readable in one color.