Proof of concept · Solo-built, end to end
Deckk: Reimagining the Link-in-Bio with AI
A market-sized category with a tired pattern: flat link lists that never tell you what matters now, and a posting flow slow enough that pages go stale. Deckk replaces both with hierarchy that understands time and a capture-to-card flow that meets creators where they already work—in a photo.
Kiln City · public deck in dark mode — Home and Events
The business problem
The link-in-bio category is no longer a niche. Linktree alone has surpassed 50 million users, and Linktree users are estimated to be driving upwards of $6 billion a year in commerce through their pages. The creator and small-business economy behind that number is enormous: over 200 million people worldwide now create and sell online, and the average person maintains 8.6 social accounts, up from 4.8 just six years ago. Every one of those accounts needs a hub.
But the category’s growth has outpaced its design. Linktree’s own data shows the strain: since 2022, its users have posted more than 100 million links pointing to 7.5 million different destinations on a single page format that has not fundamentally changed since 2016 — a flat, undifferentiated list. The product that solved “too many links” never solved “which link matters right now,” and it never solved how a business owner with no time is supposed to keep updating it.
That gap has a direct cost. A class that doesn’t fill. A pop-up with light foot traffic. A new product that never gets seen because it’s link number eleven. For a solo maker or a small business, this isn’t a UX inconvenience, it’s lost revenue on work they already did.
Two structural causes
No hierarchy
Every link carries equal visual weight regardless of urgency or relevance, so time-sensitive, high-value content (an event this week, a new product) is as easy to miss as a shop link from three years ago.
High posting friction
Adding new content requires manual form entry, typically on a phone, by someone who does not have marketing or web ops as their day job. Pages go stale because updating them is a chore.
In the wild · Heavy Manners Library
A small arts venue with a rich calendar is exactly who a hub should help. On Instagram, attention is already on them—then the bio sends everyone to Linktree. There, tonight’s poetry reading, Saturday’s workshop, last month’s press, and a permanent storefront all land as the same row in a long scroll. Nothing tells a visitor which link is worth their next hour of attention.
Which one should I tap first?
From Instagram, one bio link dumps people into the same undifferentiated stack: workshops, nights this week, shop, and press all compete as equal rows. A visitor who only has a minute may leave before they find the event that actually fills the room.
Public profile used as a product reference — not a critique of the venue’s programming.
The approach
Deckk’s core bet is that the fix isn’t a better link list. It’s a page that behaves like it knows what today is, and a posting flow fast enough that a business owner actually uses it.
Where AI sits, and why
AI is placed at the single point of highest friction in the existing workflow — manual data entry — not layered on as an add-on feature. A creator photographs a flyer or product they were making anyway for Instagram. Deckk extracts the title, date, location, price, and call to action from that image, and returns each field with a confidence score. Low-confidence fields are flagged for a quick check. Nothing publishes without the creator’s review.
This is a deliberate product decision, not a default: AI should shorten the distance between something someone already knows and the moment it goes live, and it should never remove a person from the decision to publish. That’s the design principle the confidence-scoring and mandatory review step encode.
Hierarchy, independent of AI
A hierarchy engine ranks content automatically. One pinned slot holds the creator’s stated priority. Events rise as their date approaches and step aside once it passes. New items surface briefly, then settle. The page stays current without the creator managing it.
Product · Kiln City as the reference deck
To show both sides of the system, I use Kiln City — a ceramics shop and studio — as the working example: Studio turns a flyer into structured fields, and the public deck shows how typed cards rank and display.
Studio · easy parsing with confidence
Upload a flyer. Extraction returns a draft card (title, when, where, price, CTA) with per-field confidence. Low scores ask for a human pass; high scores stay out of the way.
Public deck · content with type and priority
On deckk.me/kilncity the same content is no longer a flat list. Pinned work leads. Events know their dates. Shop items and social links settle into the shelf. Hierarchy does the curating.
Kiln City · public deck
Hierarchy and typed content for visitors — Home and Events.
Current status: proof of concept
Deckk is built and functioning, not a concept file. The hierarchy engine, the quick-post dashboard, the public page, and a first working version of the flyer-to-card extraction pipeline are live in a production Next.js and Supabase app on Vercel, with extraction running on the Anthropic API.
What that proves right now: the architecture works, the extraction pipeline reliably returns structured, confidence-scored data from an unstructured image, and the end-to-end flow from photo to published card functions.
What it does not yet prove: real-world adoption and outcome data. That’s the deliberate next phase, not a gap I’m glossing over. The validation plan:
- Moderated usability testing — task-based comparison of finding a dated event on a Deckk page versus a matched Linktree page, and of posting a new event from a flyer on both.
- Extraction accuracy tracking — logging the edit distance between what the model proposes and what the creator actually publishes, per field, to quantify real accuracy rather than estimate it.
- My own usage as the first data source — Having been a craftsperson and maker on Etsy, I am testing with my own products as well as friends in my community who run small businesses. The more I share this, the more I'm asked to beta test.
What’s next
Phase 1: Stripe Connect, so a product or event ticket sells directly on the card a visitor is already looking at, instead of routing them off-page and losing them in the hop.
Phase 2: An opt-in discovery layer so individual Deckk pages become findable by craft and neighborhood.
Why this belongs in a portfolio
This project shows product judgment applied without a brief: a market-sized problem, a mechanism-level answer for where and why AI belongs in the workflow, and production code to back it, at a stage I can speak to honestly rather than overstate.
Sources: Linktree user and revenue figures via TechCrunch (2024) and Linktree’s own reporting; creator economy figures via Contrary Research and industry aggregates, 2026.