AI Collectible Verification Platform

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 "Collectible Verification": a guided photo check that scores a resale sneaker for counterfeit risk, with a human reviewer who signs the final report.
   
   Stack: Next.js (App Router), TypeScript, Tailwind, Supabase (Postgres, Row Level Security, private Storage, Auth with email magic link), Claude (Anthropic API) with image input for the risk scan. Deploy on Vercel.
   Scope: one narrow, photographable category, sneakers.
   
   Tables (Row Level Security on: a user sees their own submissions and reports, reviewers see the queue):
   - submissions(id, user_id, brand, model, purchase_context, status, model_version, prompt_version, confidence, missing_evidence jsonb, final_status)
   - photos(id, submission_id, view, storage_path)
   - assessments(id, submission_id, raw_output jsonb, created_at)
   - reviewer_decisions(id, submission_id, reviewer_id, decision, notes, created_at)
   - reports(id, submission_id, public_slug, content jsonb, created_at)
   Storage: one private bucket with signed URLs. Strip image metadata on upload.
   
   Screens: /login, /submit (guided upload), /review (reviewer queue).
   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 single-check payments, reviewer payouts, embeddings or reference matching beyond the model's own notes, disputes, other categories.
   Done when: npm run dev starts, you can sign in, and the five tables and the private bucket exist with Row Level Security on.

2. 2. Core Feature

   Build the one feature that proves the idea: a guided photo check, a risk scan, a human sign-off and a shareable report.
   1. /submit asks for the required views: box label, size tag, outsole, stitching, logo, insole and receipt context. Save each photo to private storage.
   2. Send only the approved images to Claude with a strict JSON schema: feature_observations, inconsistencies, missing_views, reference_matches, confidence and escalation_reason. Store the raw output in assessments, separate from anything a person approved.
   3. If required evidence is missing, return a request for another photo, never a pass or fail. Show every red flag with the image region or evidence source that triggered it.
   4. /review lists submissions sorted by value, low confidence and dispute risk. A reviewer sees the photos and the model's observations, and chooses approve, reject or request more evidence.
   5. On approval, create a report with a public_slug showing the item, the evidence received, the findings, the uncertainty, the reviewer's name and the time.
   Rules: never write "guaranteed authentic". Keep raw model output and the human-approved report in separate records.
   Empty state: a new user sees the list of required photos and the upload button.
   Done when: a test pair with a missing outsole photo asks for that photo, a complete set produces an assessment, and a reviewer's approval produces a report page at /r/[slug].

3. 3. Landing Page

   Build a one-page landing site for Collectible Verification, a guided check for resale goods.
   Hero: "Know what you are buying before it ships." One button: Join the waitlist.
   Sections: the problem (counterfeits and thin evidence), how it works in four steps (submit the item, run the risk scan, get the report, escalate when needed), a sample report with red flags and an uncertainty line, and an FAQ that says plainly the report shows risk and evidence and is never a guarantee.
   Waitlist: store the email in a waitlist table in Supabase. No other service.
   Voice: careful and precise, never overclaiming.
   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 Collectible Verification: a wordmark and an icon that suggest a magnifier over a stitch, careful and not alarming.
   Colors: near-black, warm off-white and one muted green for a cleared item. Type: a precise sans for the interface and a mono for report numbers.
   Deliverables: wordmark, icon, a report header, and one launch graphic.
   Done when: each deliverable is saved in one folder and the icon reads at 32 px.