hypsh

    Data enrichment & feed quality for fashion retail

    Your product feed, finally fluent in fashion.

    hypsh reads every photo and description in your feed, turns them into structured fashion attributes, catches the feed errors that cost you sales, and gives search engines and AI shopping agents a catalogue they can actually reason about.

    fashion products enriched and live
    91,000+
    structured values per product
    Ø 72
    brands across 16 retail feeds
    2,100+

    One feed row, before and after

    A real product from a large DACH retailer's feed, brand removed. The facts are in there, somewhere: buried in run-on prose, behind a placeholder colour name, next to three price columns that disagree.

    What the feed sends

    Quoted verbatim, including the stray HTML entity.

    title
    ████ Midikleid ████ beige | 42
    product_type
    Damen > Bekleidung > Kleider > Midikleider
    colour
    beige
    description
    Das ████ Wickelkleid aus elegantem Fischgrät-Leinen-Mix beeindruckt mit oversized Revers und voluminösem Rock. Verstellbarer Bindegürtel. Zweireihiges Damen Wickelkleid aus Leinen-Mischung Breiter Umlegekragen Schließt doppelreihig mit Knöpfen Ohne Arm Bindegürtel in der Taille Seitliche Eingrifftaschen Ausgestellte Passform Lange Seitenschlitze Fischgratmuster Länge: Ca. 121cm Farbbezeichnung: Open Miscellaneous Material: 67% Leinen, 23% Polyester, 10% Viskose  
    gender
    female
    size
    42
    price
    549.00
    sale_price
    328.99
    product_price_old
    328.99

    What hypsh makes of it

    Herringbone linen-blend midi dress

    Sleeveless midi dress in a refined linen blend with a herringbone weave and an oversized lapel collar. Double-breasted buttons and an adjustable tie waist give it a defined, tailored silhouette.

    Garment
    DressMidi · 121 cmA-lineSleeveless
    Details
    Lapel collarDouble-breastedTie waistSide pocketsSplit hem
    Material
    67% linen23% polyester10% viscoseWoven twillMidweight
    Colour
    Light beigeWarmMutedNeutrals
    Pattern & feel
    HerringboneMatteCrispBreathable
    Season
    SpringSummerTransitional
    Occasion
    OfficeSpecial occasionEveningSmart / cocktail
    Flatters
    HourglassPearRectangleAccentuates waist
    Style
    ClassicMinimalistQuiet luxury
    Photos
    Ghost mannequinOn model, full bodyBack viewFabric close-up

    Real enrichment output, abridged. 13 distinct facts spread over 47 filled columns became 89 structured values, all 10 photos were classified by shot type, and the record was translated into German and French. The placeholder colour name was ignored; the colour comes from the photos.

    Feed errors, caught before your shoppers find them

    Enrichment compares what your feed says with what your photos show. Where they disagree, we fix what we can and tell you about the rest. These are real cases from live partner feeds.

    • The photos show a different garment

      The feed saidCrew-neck cashmere sweater, in the title and the description.

      ActuallyEvery photo shows a turtleneck.

      Hidden from the shop until it is fixed, so nobody orders the wrong item. Reported with a suggested correction.

    • Photos from the wrong colourway

      The feed saidOvershirt in grey.

      ActuallyThe product shot was grey, but all five on-model photos showed the brown version.

      The five wrong photos were removed from display automatically. Your original images are never edited.

    • Colour field contradicts the photo

      The feed saidColour: brown. Same row, the vendor's own colour name: “Stormy Blue”.

      ActuallyThe garment is blue.

      Search and outfit matching use the colour in the photo. The mismatch goes into your report.

    • One gallery for every colour

      The feed saidSweatshirt in black and in white, with identical photos and “black” in both titles.

      ActuallyA shopper choosing white sees a black garment. At one denim retailer this affected 14% of the range.

      Detected on every import and reported as one card per affected colour group.

    • Composition that adds up to 200%

      The feed saidMaterial: 59% viscose, 37% lyocell, 4% elastane. Lining: 64% viscose, 36% polyester.

      ActuallyRead flat, that is 200%, and the lining's viscose would earn the blazer a sustainability claim it does not deserve.

      Shell and lining are separated; lining fibres never drive material search or sustainability tags.

    • Five price columns, one right answer

      The feed saidprice, sale_price, search_price, rrp_price, product_price_old: all plausible names.

      ActuallyIn one affiliate feed the obvious column held the pre-sale price for 76% of products.

      Before go-live we check every price column against the price on your own product page.

    Your feed-quality report

    Every partner gets a private report that updates after each nightly import: what is wrong, why it matters, and what to change. In German, English or French.

    • Two-stage checking: a second, more careful AI pass re-examines every flag at full image resolution. Most first-pass suspicions are thrown out, so only confirmed problems reach you.
    • One click to remove a wrong image from display, or to mark a finding as fine and show the product as it is.
    • Self-healing: fix the item in your feed and it is re-checked on the next import, with no ticket to file.
    • Feed gaps listed in full, with CSV export: missing categories, gender, brand or images, broken image links, non-fashion items.

    What AI shopping agents need to recommend your products

    ChatGPT, Gemini and Perplexity are becoming the first place people ask what to wear. An agent can only recommend what it can reason about, and a raw feed gives it keywords, not reasons.

    I need a breathable linen midi dress for a summer wedding. I'm pear-shaped.

    This herringbone linen-blend midi (67% linen) is breathable and made for spring and summer, smart enough for a special occasion, and its A-line skirt with a tie waist flatters a pear shape by accentuating the waist. One thing to check: it is a light, warm beige, so close to bridal tones.

    From the raw feed, an agent can match “linen” and “midi” as keywords. Season, occasion, formality, body shape and the real colour are not in it.

    What the agent reads

    • linen, and how muchmaterials: linen 67%
    • breathablefabric_properties
    • midilength: midi · 121 cm
    • summerseasons: summer
    • weddinguse_contexts · formality
    • pear-shapedbody_shape_flattery: pear
    • not too close to whitecolour: light beige, warm

    • Ready for AI agents

      Structured, consistent attributes that ChatGPT apps and other agents can query directly. hypsh already runs a fashion app inside ChatGPT on this data.

    • Search that understands intent

      Shoppers can search for “something for a summer wedding” and find it, because the occasion is now in the data, not just the words in the title.

    • SEO and GEO

      Clean product names, real descriptions and translations that search engines and AI answer engines can quote, instead of keyword runs.

    How it works

    1. Connect your feed

      Any CSV or TSV product feed by URL, including Google Shopping and affiliate formats. We map your columns once.

    2. Clean and check

      Variants grouped, sizes converted, prices verified, non-fashion items removed, every row validated on each nightly import.

    3. Enrich

      A vision model reads your photos and text together and writes around 70 structured attributes per product. Where photo and text disagree, the photo wins.

    4. Use it everywhere

      In hypsh search, the outfit builder and your own ChatGPT app, or via API. Your quality report updates after every import.

    Questions

    Which feed formats do you support?

    Any CSV or TSV product feed available by URL, including Google Shopping and affiliate network formats. We set up the column mapping with you once; after that the feed is imported every night.

    Do you change our product photos?

    No. Your photos are never edited or watermarked. If an image is wrong, for example from another colourway, it is only hidden from display on hypsh, and you can restore it with one click.

    How accurate is the enrichment?

    Attributes come from your photos and your text together; where they disagree, the photo wins. Quality flags go through a second, more careful AI check at full resolution before they are reported, so you only see confirmed findings.

    Which languages?

    Enrichment is written in English and translated. German is available for the full catalogue, French and further European languages on request.

    What happens when we fix something in our feed?

    The next nightly import notices the change and re-checks the product automatically. If the problem is gone, the finding closes and a hidden product goes live again.

    See what your feed is hiding

    Send us your feed URL and we will show you an enriched sample of your own products and the issues we find.

    • Enriched sample of your own products
    • Feed-quality findings with suggested fixes
    • German, English or French

    We only use your details to reply to this request.