FullAura

I live in Vienna, and when I want to buy something new, I spend too much time walking between shops and trying things on to work out what suits me. FullAura is meant to save that time and to give people a look close to what a stylist might pick. You say what you need in your own words, choose a whole outfit, the upper body or one item, pick a clothing line, and set the most you will spend. Muse Spark 1.3, as a shopping agent, searches using German words, because the three shops that deliver here (Tom Tailor, Tchibo and Giesswein) use them in their product addresses. Code answers from the shops' sitemaps, about 44,000 addresses, and reads price and stock from each page. The model can choose only among products that code returned: in stock and, where the page gives comparable sizes, available in your size. Before you see a result, code checks the addresses, the budget, that the items make up what you asked for, and that the shop's title does not contradict the item's place in the outfit. You can then try the result on. Muse Image draws you in the clothes from two photos and each shop's title and color. It gets no photo of the product, so the drawing is approximate and says nothing about how an item looks or fits. FullAura keeps photos only in memory (for the request), and sends them to Meta in Standard mode, where Meta does not train its models on them but may retain them under its terms. Built with Muse Spark Open: a second mode plans three looks from a catalog of 37 real products, and code checks each plan against seven hard constraints. FullAura is a closed beta and states its limits on the page. Whether an item suits the occasion, warmth, whether items go together, and style are the agent's opinion. Code compares sizes only at Tom Tailor; at Tchibo and Giesswein the result says that the size was not checked. The first participants used the deployed site on 3 November, and what they met changed the product: the page now collects a search's answer when the connection drops, each result links to the shop under a clear label, and the try-on panel says that clothes you did not ask for stay as in your photo. This small test does not show usefulness. Everything in the repository was built during the event, by one person working with AI coding agents. The first commit is from 26 October 2026 at 20:34 Vienna time, after the event opened at 18:00. Meta's models and the open-source libraries were not built here. The parts: https://github.com/vbabenko97/fullaura#built-during-the-hackathon Live site (invite required, because each request is a paid model call): https://fullaura-demo.blackbay-aaeee2c5.francecentral.azurecontainerapps.io Recorded run, no invite: https://github.com/vbabenko97/fullaura/blob/main/docs/DEMO.md Local run on scripted data, no invite or keys, no model output: https://github.com/vbabenko97/fullaura#without-keys-on-scripted-data Local run with your own API keys: https://github.com/vbabenko97/fullaura#run-locally