NexTurn

NexTurn is a direct Customer-to-Customer second-life commerce prototype for returned and underused products. The core product rule is simple: no warehouse touches the item. The seller keeps the product at home until a buyer joins the queue and pickup verification unlocks payment; Amazon facilitates local pickup, quality check, and delivery.

Customer Problem

Customers lose value when usable items are returned, discarded, or pushed into opaque liquidation channels. Buyers also hesitate to trust second-hand products because condition and authenticity are unclear. NexTurn solves this from both sides of one unified account:

• sellers list real products from verified order history;
• buyers inspect order proof, preliminary AI comparison, purchase date, seller location, and estimated delivery fee before joining the buyer queue;
• Amazon acts as the trust and logistics layer, not as a warehouse middle step.

Implemented Prototype

• Unified Buyer + Seller account shell with Cognito Hosted UI and optional Google sign-in.
• Sell / Return Items hub gated behind login.
• Amazon order proof history with 5 high-value products, original price, purchase date, ASIN, original product image, and proof metadata.
• Real upload flow for seller item photos.
• AWS Rekognition image evidence on deployed API, combined with deterministic photo comparison for product identity, order-photo similarity, dominant colour/variant, and visible damage risk.
• Damage-aware grading: cracked or broken screen evidence forces low grades like C, and variant or colour mismatches cannot receive an A.
• Dynamic preliminary resale value based on the AI comparison result.
• Address-gated accounts: every signed-in account must save an India delivery address before listing an item or joining a buyer queue.
• Global C2C marketplace that merges NexTurn AI-graded listings with 100+ public API background products from DummyJSON.
• Listing detail drawer with transparent order proof, purchase date, AI comparison result, original product image, seller-upload image, preliminary price, delivery estimate, and "AI Graded & Amazon Verified" badge.
• Search, category, grade, queue-status, and sort controls for marketplace browsing.
• Buyer queue flow: buyers express interest, payment remains locked, and the final value is confirmed only after manual pickup inspection. A listing queue becomes filled after the first buyer joins.
• Green credits for sustainable C2C behavior: sellers earn credits for listing order-proof-backed second-life items, and buyers earn pending credits after pickup review confirms reuse.
• DynamoDB persistence for customer profiles, created listings, and buyer queue interest records.
• S3 media persistence and Rekognition permissions in the AWS CDK stack.

Stage 2 AI Hooks

These are documented as expansion hooks in the current architecture, not as customer-facing payment or logistics automation in this prototype:

• Feature label: Predictive Return Prevention (Before purchase is made)
• Feature label: Multi-Routing AI (Automatically deciding if an item goes to Refurbished, Donated, or Recycled streams if it fails C2C standards)
• Predictive Return Prevention (Before purchase is made): before checkout, NexTurn can compare a shopper's fit, size, variant, return, and preference history against the current product metadata to warn about likely mismatch and suggest a safer second-life or exchange option.
• Multi-Routing AI (Automatically deciding if an item goes to Refurbished, Donated, or Recycled streams if it fails C2C standards): if the image proof, confidence score, damage evidence, or pickup review fails the direct C2C threshold, the same grading policy can route the item to an alternative stream instead of listing it in the marketplace.

Deployed Screenshots

Captured from the live AWS deployment: https://l5f3ovamaj.execute-api.us-east-1.amazonaws.com/

Overview

Second-life shop

Listing proof drawer

AI grading outcomes

The return scan flow blocks wrong-product uploads, downgrades variant mismatches, and keeps every grade preliminary until pickup review.

Architecture

flowchart LR
  Seller["Customer A: seller"] --> Web["React NexTurn app"]
  Buyer["Customer B: buyer"] --> Web
  Web --> Auth["Cognito Hosted UI + Google"]
  Web --> API["HTTP API Gateway"]
  API --> JWT["JWT authorizer for buy/sell"]
  API --> Lambda["NexTurn Lambda"]
  Lambda --> Orders["Amazon order proof history"]
  Lambda --> Rekognition["Amazon Rekognition"]
  Lambda --> S3["Private S3 scan media"]
  Lambda --> Grading["AI comparison + preliminary price engine"]
  Lambda --> DDB["DynamoDB profiles, listings + buyer queues"]
  Lambda --> PublicFeed["DummyJSON public marketplace feed"]
  DDB --> Marketplace["Global C2C marketplace"]

Loading

Main Flow

1. Seller signs in.
2. Seller saves an India delivery address in Profile.
3. Seller opens Sell / Return Items and selects a product from Amazon order history.
4. Seller uploads a real item photo.
5. Backend compares the upload against the order proof photo, metadata, and expected colour/variant, then calculates a preliminary grade and resale value.
6. Seller publishes the listing. The item stays with the seller at home.
7. Another signed-in buyer saves an India delivery address, opens Marketplace, and sees the listing above the public API product feed.
8. Buyer opens the listing, checks order proof, original purchase date, preliminary AI comparison, price, seller location, and estimated delivery fee.
9. Buyer joins the buyer queue. The listing becomes queue-filled for other users, payment stays locked, and green credits remain pending until an Amazon delivery partner verifies the item at pickup and opens the final payment step.

Local Development

npm install
npm run dev

Open http://127.0.0.1:5173/.

Local mode can render the app and marketplace. Authenticated buy/sell flows are fully available on the deployed AWS URL because Cognito provides real JWTs.

Verification

npm run test
npm run smoke:api
npm run build
npm run cdk:synth

AWS Deployment

The CDK stack targets us-east-1 by default and uses Free Tier friendly services: HTTP API, Lambda, DynamoDB on-demand, S3, Cognito, and Rekognition.

npm run cdk:synth
npm run cdk:deploy

Google sign-in is configuration-gated. Set GOOGLE_CLIENT_ID and GOOGLE_CLIENT_SECRET before deploy to keep Google federation enabled.

Key Files

• src/App.jsx - C2C buyer/seller web app.
• src/data/c2cCommerce.js - order proof history, demo accounts, seed listings, and supported India delivery locations.
• src/lib/c2cCommerce.js - product comparison, grading, pricing, address, delivery, listing, and buyer queue rules.
• backend/lambda/returnResolution.js - Lambda-compatible API handler.
• backend/lib/aiImageAnalysis.js - S3 upload and Amazon Rekognition analysis.
• backend/lib/dynamodbRepository.js - DynamoDB persistence adapter.
• infra/cdk/app.mjs - AWS infrastructure.