Giftmaxxing 🎁

Gift-giving is a craft.
Giftmaxxing learns the taste of the people you love, remembers every date that matters, and turns “I have no idea what to get them” into the perfect gift — on your own or as a group.

🌐 Live app  ·  The swipe flow  ·  Feature tour  ·  Under the hood

---

Three gifts, up top

🎁 Know their taste Swipe a handful of finds and Giftmaxxing learns what someone actually wants — no awkward asking.

🎁 Never forget a date Birthdays, milestones and shared occasions — tracked, with reminders and ideas ready before the day.

🎁 Gift-giving is a craft A taste-ranked feed, an AI concierge and curated drops help you give something thoughtful, every time.

---

✨ Learn their taste by swiping

The signature flow. Swipe right on what fits, left on what doesn’t — every “yes” sharpens a taste vector that powers your gift matches. Send the challenge to a friend and a soft profile of their taste lands in your account (with their okay), so you’re never guessing again.

_{Five “wants” unlock a ranked set of gift matches, blended from your swipe taste.}

---

🧭 Feature tour

Everything Giftmaxxing does, section by section.

📅 Never forget a date — Events & milestones

Personal and shared timelines for birthdays, anniversaries and goals. Each milestone carries a reward and surfaces gift ideas as the date approaches.

👥 Group gifts that actually land

Pool money toward one great gift — everyone chips in, nobody double-buys. Live progress bars, contributor avatars, and an Invite button to bring people in (in-app or via a share link, with a clear no-payments-handled disclaimer).

🤖 Maxi — your AI gift concierge

Tell Maxi a budget, a vibe, or who it’s for. It finds the gift, adds it to your cart, and can even check out for you.

🏠 A feed of giftable finds

The home surface: a taste-ranked feed of gift-worthy finds, group pitch-in chats, people to follow, and Maxi nudges (“their birthday is in 4 days — I lined up 7 ideas in your budget”).

🎀 Curated drops

Hand-assembled gift bundles — themed, refreshed, and ready to gift together or add to cart in one tap.

🛍️ Shop

Hundreds of hand-picked gifts on Amazon. Affiliate-supported, with a clear Associates disclosure — prices and availability shown on Amazon.

🙋 Your gifting profile

A profile built from your onboarding taste — your gifting role, style and budget, the things you’re into, a public/private toggle, and your posts + friends (soft profiles you’ve collected). No vanity follower counts.

🔔 Activity

Everything happening around your gifting — reward unlocks, price drops on your radar, pool contributions, claims from your wishlist, and new followers.

🧪 Recommendations Lab

A side-by-side demo of the recommender: pick a persona and compare the client-side facet ranker against the server-side S3 Vectors recommender, with the computed taste vector visualized.

🚀 Onboarding

A short wizard captures the taste signals that feed every recommendation — name, gifting role, style, budget and interests.

---

📱 Built for mobile, too

Swipe

Events

Feed

Group gifts

Screenshots and the GIF are generated with Playwright against the local app (web/). To refresh them, run the dev server and re-run the capture script.

---

Monorepo layout

Path	What	Deploys to
web/	Next.js app (feed, recommender UI, landing)	Vercel (GitHub auto-deploy on push to main; Root Directory = web)
infra/	Terraform serverless backend: DynamoDB + Lambda + API Gateway HTTP API	AWS us-east-1 (deployed)
infra/ingest/	Reddit → DynamoDB data-loading scripts	run locally

The web app calls the AWS API via NEXT_PUBLIC_API_URL.

Specs & roadmap (read these first)

Doc	Purpose
CLOUD.md	Canonical cloud/AWS spec + agent memory for the Pinterest-image → multimodal-embedding → vector-index → recommendation pipeline, the Pinterest-style native-ad simulation, cost analysis, and the future visual-search feature. Agents: treat CLOUD.md §8 as the working backlog.
infra/README.md	Terraform resources, deploy steps, API routes, planned embedding pipeline.
web/README.md	Next.js app dev instructions.

Image → vector → feed pipeline (summary)

Connected Pinterest boards/pins (taste signal, user-scoped) and our catalog → Amazon S3 → Amazon Bedrock Titan Multimodal Embeddings (unified text+image vectors) → vector index (brute-force in Lambda now → Amazon S3 Vectors at scale) → blended into the recommendation feed by cosine similarity. Sponsored items are interleaved seamlessly (same card, subtle "Sponsored" label, ranked by the same taste vector), mirroring Pinterest's native-ad pattern. Full design, rate limits, and cost (≈ a few $/mo at dev scale) live in CLOUD.md.

Future: visual search ("Google Lens for gifts")

Upload/snap a photo → embed with the same multimodal model → kNN against the product catalog → return visually similar buyable products with Amazon / Walmart affiliate links. Not built yet — designed and indexed in CLOUD.md §6 for a later agent to implement.

Getting started

• Web: see web/README.md.
• Infra: see infra/README.md.
• Copy .env.example → .env and fill in real values (never commit .env).