Nook | Devpost

Nook asking for personalization questions from user on shopping habits
Signing in page on Nook
Setting up personalization content for user
Quest page - includes personal quests and donations to the community using acorns
Avatar page - users can decorate their avatar freely and uniquely through accessories and purchases using acorns
Pip's deal checks - Pip re-checks prices 24 hours after you choose to wait after discussion with Pip
Profile card of the user
Main page - shows impact of how many impulsive purchases you prevented, the amount of money saved, conversations with Pip, decisions, etc
UI interface of Pip that shows up when user is trying to checkout
Rewards through exp of character and acorns (currency within our service) after user chooses a decision upon discussion with Pip
The different choices users can make upon discussion with Pip - Buy now, wait, and skip/save money.
Another UI of how the user sees Pip when they are trying to purchase a product

Inspiration

Smart savings. Online checkout is built to be frictionless, thereby encouraging impulse buys and regretful returns. Loyalty programs make it worse by rewarding you for spending more, and cashback apps are passive and easy to forget. We wanted one short, friendly pause right before you pay, and a reward for thinking it through instead of for spending.

What it does

Get to know Pip: after you sign in, Pip interviews you by voice about your shopping habits and goals. Gemini turns your answers into a short shopper profile you can review and edit, and Pip uses it to personalize every checkout conversation.
Checkout check-in: when you reach a checkout page on a real store, Pip pops up and talks with you by voice about your cart. You make your case, and Pip suggests buying, waiting 24 hours, or skipping.
You decide: Pip advises but never blocks. Every decision earns acorns and XP. Skipping earns the most, and following Pip’s advice earns more than going against it.
Better deals, not just less spending: if you buy, Pip can hunt for a better price before you pay. It shows cheaper sellers, coupon codes, and similar products that meet the same need for less, each with why it fits and the trade-off. If you wait, Pip checks the prices again 24 hours later and has the results ready on your dashboard.
Pip grows with you: XP moves Pip up five tiers, from Baby to Golden. You can spend acorns on outfits and pets in the avatar closet.
Weekly quests: three personal quests each week, like “Talk to Pip before 3 checkouts” or “Keep $20 from impulse buys”. The targets get harder as you complete them.
Your forest: 100 acorns plants a tree, and 30 trees completes a forest, which is a $5 donation pledge. A leaderboard ranks people by trees planted.
Impact dashboard: see what you’ve kept from impulse buys, how your decisions add up over time, and the results of Pip’s price checks.

How we built it

Chrome extension (Manifest V3, built with Vite): a content script that spots checkout pages, the Pip popup, and pages for onboarding, the dashboard, avatar closet and quests. Pip is drawn in SVG and animated to sync with the voice.
FastAPI backend that checks each Supabase login token and handles onboarding, checkout sessions, decisions, rewards, deal checks, cosmetics, quests and the forest. Live progress (like the deal hunt) streams to the extension as it happens.
Supabase for Google sign-in and the Postgres database. Rewards, quest claims and tree planting run as database functions, so balances can’t be double-spent.
ElevenLabs voice agents for the onboarding interview and the checkout conversation, plus Eleven Turbo text-to-speech for the backup voice.
Gemini turns conversations into structured data: your shopper profile, Pip’s verdict, and a purchase category for each checkout. It also powers the deal agent, which calls tools to search Google Shopping (through SerpAPI) and check coupons.

Challenges we ran into

Keeping rewards fair: we made sure an acorn reward can only be paid once and that you can’t spend more acorns than you have, even when two requests arrive at once.
Trustworthy deals: AI models can make up prices and links. The deal agent only uses offers that came back from a real search, drops used and refurbished items, and calculates savings in code instead of trusting the model’s math.

Accomplishments that we're proud of

It rewards spending well: skipping, waiting and following good advice earn more than impulse buying.
Pip feels like a companion rather than a nagging presence. It knows your goals, reacts while you talk, and grows the more you use it.

What we learned

How to get an AI model to return reliable structured data (a fixed set of JSON fields) instead of free text.
What Chrome extensions allow and block around permissions, sign-in and scripts.
How to put game-like rewards (tiers, quests, leaderboards) on a database without breaking each other’s data.

What's next for Nook

Turning donation pledges into real donations with partnerships and affiliate marketing with explicit user consent.
Making Pip’s voice even more natural and capable: faster responses, more languages, and conversations that remember your past decisions.