Aucti, an AI-native auction platform.
What if listing an item took 30 seconds and bidding felt like a conversation, not a form?
Executive summary
Aucti is a peer-to-peer auction platform I founded and design end-to-end. It uses AI to generate listings from a single photo, runs conversational bidding instead of form-based bidding, and is built to make the moment of selling something as low-friction as messaging a friend.
I lead research, concept, interaction design, brand, and product decisions, working with a small engineering team to ship.
The problem.
Marketplaces ask sellers to do unpaid labour. Photograph it, describe it, price it, categorize it, then babysit the listing. Buyers, in return, get to wade through inconsistent listings and a bidding UI optimized for the platform, not for them. Both sides churn.
If the AI can do the listing work and the conversation can do the bidding work, the platform stops being a chore and starts being something you want to open.
My role.
Founder and sole designer. Everything from concept and research through interaction, brand, and shipping.
- Defined the product premise.
Wrote the founding thesis, validated it with seller and buyer interviews, and held the line on the two non-negotiables: AI listings and conversational bidding.
- Designed every surface.
Onboarding, listing creation, bidding, payment, post-sale handoff, notification system, and the seller dashboard. Solo design ownership.
- Built the brand.
Name, voice, identity system, marketing site. The brand is opinionated because the product is.
- Shipped with engineering.
Worked directly with two engineers, doing my own prototyping, edge-case mapping, and QA. Designers who don't watch the build lose their decisions.
Unique challenges.
- AI listings have to be trustworthy.
Auto-generated listings are only an asset if sellers feel ownership of them. Designing the edit-and-approve moment was where the entire premise stood or fell.
- Conversational bidding is novel.
Users have decades of mental models for bidding boxes. Replacing them with a chat-shaped flow required heavy onboarding work and careful default behaviour.
- Solo founder, real product.
Every decision is mine and the bus factor is one. I structured the design system and component library specifically to let a future team join without re-litigating settled decisions.
My process, highlights and takeaways.
- Research before the prompt.
The AI listing generator is informed by seller interviews, not by what the model could plausibly produce. Real seller language drives the output schema.
- Conversational bidding tested in low-fi.
Paper and Figma prototypes long before the model was wired up. The novelty had to earn its place before engineering touched it.
- Trust through transparency.
Sellers see exactly what the AI generated, why, and edit it before publishing. The model is fast, but the seller is the author.
- Brand and product as one decision.
The voice, the bidding tone, the notification copy, and the visual identity all came from the same brief, written before the first screen.
Final thoughts.
What worked well. Treating AI as a way to do users' work, not the platform's, kept the experience honest. Sellers feel like the platform did something for them, not to them.
What I would change. I would have invested in the post-sale handoff earlier. Buying ends at "won." Owning starts after that, and the design of the handoff between the platform and the relationship matters more than I initially weighted.
Thanks to the early beta sellers who told me what was working and, more usefully, what wasn't.