Retrofit Uses AI to Find You Vintage Clothes That Don't Look Like a Costume

The Macro: Vintage Is Having a Moment, but Shopping for It Still Sucks

The secondhand clothing market is projected to hit $350 billion globally by 2028, according to ThredUp’s latest resale report. In the US alone, resale is growing five times faster than the broader retail clothing market. That’s not a trend. That’s a structural shift driven by sustainability concerns, price sensitivity, and the simple fact that a lot of people think older clothes look better than what’s on the rack at Zara.

But the shopping experience hasn’t kept pace with the demand. If you want to buy vintage right now, your options break down into a few buckets. There’s the in-person route: spend a Saturday digging through racks at your local thrift store or curated vintage shop. That’s fun exactly once a month and exhausting every other time. There’s the online marketplace route: Depop, Poshmark, eBay. These platforms have enormous inventory, but discovery is a nightmare. You’re scrolling through thousands of listings with inconsistent photos, unreliable sizing, and no quality guarantee. ThredUp has tried to solve this at scale with their warehouse model, but the result often feels like shopping at a digital Goodwill.

The problem is fundamentally about curation. Vintage inventory is unique by definition. Every piece is one of one. That makes traditional e-commerce approaches like category filtering, size sorting, and keyword search much less effective than they are for new goods. You can’t just search “black leather jacket size medium” and expect good results when every jacket is different.

A few companies have tried to layer AI onto this problem. Depop uses recommendations. ThredUp has search improvements. But nobody has really built the full personal shopper experience where AI understands your style, monitors incoming inventory across multiple sources, and surfaces pieces before you even know to look for them.

The Micro: A Stanford MBA Who Volunteered at Vintage Stores

Sandra Lifshits is the CEO. She previously worked in product at e-commerce startups, has an MBA from Stanford and a degree from Brown. Before starting Retrofit, she volunteered at vintage stores in New York to understand operations and shopper behavior from the inside. That’s the kind of primary research that sounds small but usually produces better product instincts than any amount of market analysis. Maddy Yip is the co-founder, bringing computer vision expertise from her time at a large tech company and Stanford CS. She led two computer vision patents focused on visual understanding, which maps directly to the challenge of evaluating and categorizing unique vintage items at scale.

Retrofit works like this: you create a style profile, and an AI personal shopper named Sophie curates vintage pieces for you. Sophie pulls from a network of vetted independent vintage stores, trusted resellers, and online marketplaces across the country. Most customers get their first picks within hours of completing their profile. Returns are free within 14 days. Items are filtered for condition and authenticity before they’re surfaced.

The company is a two-person team based in New York, part of YC’s Winter 2025 batch. They’re currently US-only and priced in the moderate range.

What I find interesting about the approach is that it’s aggregation with taste, not just aggregation. A lot of marketplace plays in secondhand have tried to win on selection. More listings, more categories, more inventory. Retrofit is betting that the right five items shown to the right person beats the right five thousand items shown to everyone. That’s a fundamentally different product philosophy.

The Verdict

I like the bet here. The vintage market has a genuine curation bottleneck, and AI personal shopping is a more natural solution than better search filters or bigger catalogs. The supply side is also interesting. Independent vintage stores have inventory they can’t move online effectively because they don’t have the time or expertise to photograph, list, and market every piece. If Retrofit can become their digital storefront with built-in demand, that’s a real two-sided network effect.

The risks are the same ones every curated marketplace faces. Can Sophie’s taste scale? Early users with clear, well-defined style preferences will probably have a good experience. The harder test is the customer who says “I don’t know what I want, show me something cool.” That’s where AI curation either shines or faceplants.

In thirty days, I’d want to see repeat purchase rates. Are people buying once and disappearing, or coming back? Sixty days, I’d want to know how many independent vintage stores are actively supplying inventory, and whether that number is growing organically. Ninety days, the question is whether Sophie’s recommendations improve with usage data or flatten out. The personal shopper metaphor only works if the shopper actually learns.