OOtira Named a Finalist in the Gemma Developer Challenge at Google I/O Connect China 2026
A weather-aware closet assistant built on Gemma 4 vision will present at the final round in Shanghai on 12 August
By OOtira Team
The OOtira team today announced that OOtira, a weather-aware closet assistant, has advanced to the final round of the Gemma Developer Challenge at Google I/O Connect China 2026. The team will present at the final round held in Shanghai on 12 August.
OOtira placed second in its track in the semi-final round.
About the product
OOtira is a closet decision tool built for the North American market. It consists of a Chrome extension and a web application.
Unlike conventional wardrobe applications, OOtira does not require users to catalogue their entire wardrobe before the product becomes useful. Its core feature, Should I Buy, operates at the point of purchase. On a retail product page, the user clicks the extension icon; the system evaluates the item against the clothes the user already owns and against the next eight days of local weather, then returns one of three verdicts — Skip, Wait, or Buy — with a written reason. Items the user chooses to keep are added to the closet automatically, so wardrobe data accumulates as a by-product of use rather than as a prerequisite for it.
The web application provides daily outfit recommendations based on local weather, and can compose the day's outfit into a magazine-style flat-lay image.
Technical implementation
OOtira uses Gemma 4 for garment recognition. Images are processed through Cloudflare AI Gateway to Workers AI for edge inference, producing structured fields including category, colour, material, warmth rating, and style. These fields feed the decision engine.
Verdicts are computed from four deterministic signals: redundancy (overlap with existing garments across category, colour, material, and warmth), gap (whether the item fills an actual absence in the wardrobe), weather (wearability across the eight-day forecast), and pairability (how many outfit combinations the item forms with garments already owned). A language model is used only to render the verdict as a written reason, and that reason is validated against the user's actual closet so that it cannot cite garments the user does not own.
The system runs on Cloudflare's edge platform across seven Workers, with D1 for the database, R2 for object storage, KV for caching, and Workers AI for inference. The browser extension performs product recognition across eight major retail sites.
Development status
Development began in late May 2026. Team members have built the product outside their primary employment, over a period of roughly two months. OOtira is currently in closed beta and has not opened public registration.
With no paid advertising, no press coverage, and content operations not yet formally underway, the project site recorded approximately 2,400 unique visitors in the past month, and 15 people have joined the waiting list.
What comes next
The team is developing a user-side closet agent. The agent will re-evaluate items on a user's wishlist daily and notify the user when the verdict changes. Planned capabilities include cost-per-wear accounting, seasonal and occasion-based anticipation, and long-term preference learning.
The team states that the agent is designed to reason on the user's behalf, and will advise against a purchase when that is the correct answer.
Links
Google I/O Connect China 2026: https://ioconnectchina.googlecnapps.cn/
Gemma Developer Challenge: https://hackathon.googdg.cn/
OOtira: https://style.ootira.com