NoGiftCards - Your Friends Deserve Better Gifts (Probably)

Stop giving gift cards, you coward

Comment

Inspiration

Most people give gift cards because they genuinely don't know what to get not because they don't care. Gift-giving is hard. It requires knowing someone deeply, understanding what they'd love but would never buy themselves, and then actually finding that thing across dozens of platforms. We've all been there: 40 tabs open at midnight, panic-buying a candle and wondering if we'll think about it.

We wanted to fix that. Not with a curated list or a quiz but with actual AI that knows the person the way their texts do, and actual agents that shop the way a personal shopper would.

The name came naturally: gifting is an art. Finding the perfect gift feels like alchemy.

What We Built

NoGiftCards is an AI-powered gift concierge that turns raw text conversations into hyper-personalized gift recommendations discovered live, scored by science, bought with confidence.

The flow:

  1. Paste Their Texts Drop in a WhatsApp or iMessage conversation. The webapp extracts personality traits, quirks, hobbies, and a full Gift DNA profile across 5 axes: Sentimental ↔ Practical ↔ Adventurous ↔ Luxurious ↔ Quirky.

  2. Watch An AI Shop Browser agents (powered by Gemini + Browser Use) deploy across Etsy, Amazon, specialty stores, and niche shops in real time, with a live adventure map showing exactly where the agent is and what it's thinking.

  3. Scored By Science Every discovered gift gets an Alchemy Score across four dimensions:

$$\text{Alchemy Score} = w_1 \cdot P_{match} + w_2 \cdot U_{factor} + w_3 \cdot B_{fit} + w_4 \cdot S_{surprise}$$ Where \( P_{match} \) is personality match, \( U_{factor} \) is uniqueness, \( B_{fit} \) is budget fit, and \( S_{surprise} \) is surprise factor.

Buy With Confidence Add favorites to cart, get direct purchase links, and ship it.

No more gift cards. No more panic-Googling.

How We Built It

Frontend React 18 + TypeScript + Vite, styled with Tailwind CSS and shadcn/ui components. We built a fully custom alchemical design system parchment cards, gold accents, Cinzel serif fonts to make the experience feel like an actual quest. The live quest dashboard includes a real-time adventure map, agent thought stream console, and a discovery grid that populates as gifts are found.

AI Profile Analysis Supabase Edge Functions parses raw conversation text and returns structured JSON: personality tags, quirks, hobbies, a prose summary, and the 5-axis Gift DNA profile rendered as a radar chart with Recharts.

Browser Agents Gemini-powered agents use Browser Use to autonomously navigate shopping sites, extract product data, and return structured gift candidates. The frontend polls quest status every 3 seconds via useQuestPolling, updating the UI in real time.

Backend Supabase handles auth, database (PostgreSQL), and serverless Edge Functions (analyze-profile, launch-quest, quest-status, redirect-quest). No dedicated server needed.

Challenges We Faced

1. Making AI output feel alive, not loading. A progress bar while AI shops feels like watching paint dry. We rebuilt the entire quest experience around visibility the agent thought stream, the adventure map, discoveries appearing one by one. Making the wait feel intentional and satisfying was harder than the AI itself.

2. Structured extraction from messy conversations. Real texts are chaotic slang, typos, inside jokes, half-sentences. Getting Claude to reliably extract clean, structured Gift DNA from noisy input required significant prompt engineering and output schema validation with Zod.

3. Real-time frontend ↔ agent sync. Browser agents run async on the backend. Bridging that into a smooth, live frontend experience without websockets meant designing a polling architecture that felt instant without hammering the DB.

4. Multi-platform agent reliability. Etsy, Amazon, and niche stores all have wildly different DOM structures, bot detection, and rate limits. Building agents resilient enough to navigate all of them consistently was the biggest technical lift of the weekend.

What We Learned

  • The user experience of waiting matters as much as the result. Make AI feel alive.
  • Supabase Edge Functions + browser agents can replace an entire backend team at hackathon speed.
  • Personality is in the subtext the things people text about reveal more than any quiz.

Built With

Share this project:

Updates