Fleek raises $25m Series B for AI‑driven secondhand fashion infra

Fleek is a good example of how applied AI is quietly eating entire operational layers in legacy industries rather than just adding a chatbot. Its custom vision-language model, trained on millions of secondhand transactions, is turning what used to be a deeply manual, tacit-knowledge-heavy task—grading used garments in sorting hubs—into a repeatable AI workflow. That matters because supply-chain domains like this are rich in messy edge cases, long tails and high economic value, exactly the environments where robust multimodal models and agents get stress‑tested.

From an AGI lens, Fleek underscores how real leverage comes from pairing models with proprietary data and domain-specific harnesses. The core model here isn’t a frontier system; it’s a targeted model wrapped in processes, feedback loops and a marketplace. As more verticals build similar “model + marketplace” stacks, they’ll generate hard-to-replicate datasets on how AI decisions map to real‑world outcomes. Those data streams are the raw material for more general reasoning systems, and the commercial pull for making frontier models better at long-horizon, high-noise industrial tasks.